Stikked
HAProxy
HAProxy
Configuration
----------------------
----------------------
version 1.4.
willy
2012/08/14
2012/08/14
This document covers the configuration language as implemented in the
specified above. It does not provide any hint, example or advice. For
documentation, please refer to the Reference Manual or the Architecture
The summary below is meant to help you search sections by name and
through the
Note to documentation contributors
This document is formated with 80 columns per line, with even number
spaces for indentation and without tabs. Please follow these rules
so that it remains easily printable everywhere. If a line needs to
printed verbatim and does not fit, please end each line with a
('\') and continue on next line. If you add sections, please update
summary below for easier
Summary
-------
Summary
-------
1. Quick reminder about
1.1. The HTTP transaction
1.2. HTTP
1.2.1. The Request
1.2.2. The request
1.3. HTTP
1.3.1. The Response
1.3.2. The response
2. Configuring
2.1. Configuration file
2.2. Time
2.3.
3. Global
3.1. Process management and
3.2. Performance
3.3.
3.4.
4.
4.1. Proxy keywords
4.2. Alphabetically sorted keywords
5. Server and default-server
6. HTTP header
7. Using ACLs and pattern
7.1. Matching
7.2. Matching
7.3. Matching regular expressions
7.4. Matching IPv4
7.5. Available matching
7.5.1. Matching at Layer 4 and
7.5.2. Matching contents at Layer
7.5.3. Matching at Layer
7.6. Pre-defined
7.7. Using ACLs to form
7.8. Pattern
8.
8.1. Log
8.2. Log
8.2.1. Default log
8.2.2. TCP log
8.2.3. HTTP log
8.3. Advanced logging
8.3.1. Disabling logging of external
8.3.2. Logging before waiting for the session to
8.3.3. Raising log level upon
8.3.4. Disabling logging of successful
8.4. Timing
8.5. Session state at
8.6. Non-printable
8.7. Capturing HTTP
8.8. Capturing HTTP
8.9. Examples of
9. Statistics and
9.1. CSV
9.2. Unix Socket
1. Quick reminder about
----------------------------
----------------------------
When haproxy is running in HTTP mode, both the request and the response
fully analyzed and indexed, thus it becomes possible to build matching
on almost anything found in the
However, it is important to understand how HTTP requests and responses
formed, and how HAProxy decomposes them. It will then become easier to
correct rules and to debug existing
1.1. The HTTP transaction
-------------------------------
-------------------------------
The HTTP protocol is transaction-driven. This means that each request will
to one and only one response. Traditionally, a TCP connection is
from the client to the server, a request is sent by the client on
connection, the server responds and the connection is closed. A new
will involve a new connection
[CON1] [REQ1] ... [RESP1] [CLO1] [CON2] [REQ2] ... [RESP2]
In this mode, called the "HTTP close" mode, there are as many
establishments as there are HTTP transactions. Since the connection is
by the server after the response, the client does not need to know the
length.
length.
Due to the transactional nature of the protocol, it was possible to improve
to avoid closing a connection between two subsequent transactions. In this
however, it is mandatory that the server indicates the content length for
response so that the client does not wait indefinitely. For this, a
header is used: "Content-length". This mode is called the "keep-alive" mode
[CON] [REQ1] ... [RESP1] [REQ2] ... [RESP2]
Its advantages are a reduced latency between transactions, and less
power required on the server side. It is generally better than the close
but not always because the clients often limit their concurrent connections
a smaller
A last improvement in the communications is the pipelining mode. It still
keep-alive, but the client does not wait for the first response to send
second request. This is useful for fetching large number of images composing
page
[CON] [REQ1] [REQ2] ... [RESP1] [RESP2]
This can obviously have a tremendous benefit on performance because the
latency is eliminated between subsequent requests. Many HTTP agents do
correctly support pipelining since there is no way to associate a response
the corresponding request in HTTP. For this reason, it is mandatory for
server to reply in the exact same order as the requests were
By default HAProxy operates in a tunnel-like mode with regards to
connections: for each connection it processes the first request and
everything else (including additional requests) to selected server.
established, the connection is persisted both on the client and
sides. Use "option http-server-close" to preserve client persistent
while handling every incoming request individually, dispatching them one
another to servers, in HTTP close mode. Use "option httpclose" to switch
sides to HTTP close mode. "option forceclose" and
http-pretend-keepalive" help working around servers misbehaving in HTTP
mode.
mode.
1.2. HTTP
-----------------
-----------------
First, let's consider this HTTP request
Line
number
number
1 GET /serv/login.php?lang=en&profile=2 HTTP/1.
2 Host: www.mydomain.
3 User-agent: my small
4 Accept: image/jpeg,
5 Accept:
1.2.1. The Request
-----------------------
-----------------------
Line 1 is the "request line". It is always composed of 3 fields
- a METHOD :
- a URI : /serv/login.
- a version tag : HTTP/1.
All of them are delimited by what the standard calls LWS (linear white
which are commonly spaces, but can also be tabs or line feeds/carriage
followed by spaces/tabs. The method itself cannot contain any colon (':')
is limited to alphabetic letters. All those various combinations make
desirable that HAProxy performs the splitting itself rather than leaving it
the user to write a complex or inaccurate regular
The URI itself can have several forms
- A "relative URI"
/serv/login.
It is a complete URL without the host part. This is generally what
received by servers, reverse proxies and transparent
- An "absolute URI", also called a "URL"
http://192.168.0.12:8080/serv/login.
It is composed of a "scheme" (the protocol name followed by '://'), a
name or address, optionally a colon (':') followed by a port number,
a relative URI beginning at the first slash ('/') after the address
This is generally what proxies receive, but a server supporting HTTP/1.
must accept this form
- a star ('*') : this form is only accepted in association with the
method and is not relayable. It is used to inquiry a next
capabilities.
capabilities.
- an address:port combination : 192.168.0.
This is used with the CONNECT method, which is used to establish
tunnels through HTTP proxies, generally for HTTPS, but sometimes
other protocols
In a relative URI, two sub-parts are identified. The part before the
mark is called the "path". It is typically the relative path to static
on the server. The part after the question mark is called the "query
It is mostly used with GET requests sent to dynamic scripts and is
specific to the language, framework or application in
1.2.2. The request
--------------------------
--------------------------
The headers start at the second line. They are composed of a name at
beginning of the line, immediately followed by a colon (':').
an LWS is added after the colon but that's not required. Then come the
Multiple identical headers may be folded into one single line, delimiting
values with commas, provided that their order is respected. This is
encountered in the "Cookie:" field. A header may span over multiple lines
the subsequent lines begin with an LWS. In the example in 1.2, lines 4 and
define a total of 3 values for the "Accept:"
Contrary to a common mis-conception, header names are not case-sensitive,
their values are not either if they refer to other header names (such as
"Connection:"
The end of the headers is indicated by the first empty line. People often
that it's a double line feed, which is not exact, even if a double line
is one valid form of empty
Fortunately, HAProxy takes care of all these complex combinations when
headers, checking values and counting them, so there is no reason to
about the way they could be written, but it is important not to accuse
application of being buggy if it does unusual, valid
Important
As suggested by RFC2616, HAProxy normalizes headers by replacing line
in the middle of headers by LWS in order to join multi-line headers.
is necessary for proper analysis and helps less capable HTTP parsers to
correctly and not to be fooled by such complex
1.3. HTTP
------------------
------------------
An HTTP response looks very much like an HTTP request. Both are called
messages. Let's consider this HTTP response
Line
number
number
1 HTTP/1.1 200
2 Content-length:
3 Content-Type:
As a special case, HTTP supports so called "Informational responses" as
codes 1xx. These messages are special in that they don't convey any part of
response, they're just used as sort of a signaling message to ask a client
continue to post its request for instance. In the case of a status 100
the requested information will be carried by the next non-100 response
following the informational one. This implies that multiple responses may
sent to a single request, and that this only works when keep-alive is
(1xx messages are HTTP/1.1 only). HAProxy handles these messages and is able
correctly forward and skip them, and only process the next non-100 response.
such, these messages are neither logged nor transformed, unless
state otherwise. Status 101 messages indicate that the protocol is
over the same connection and that haproxy must switch to tunnel mode, just
if a CONNECT had occurred. Then the Upgrade header would contain
information about the type of protocol the connection is switching
1.3.1. The Response
------------------------
------------------------
Line 1 is the "response line". It is always composed of 3 fields
- a version tag : HTTP/1.
- a status code :
- a reason :
The status code is always 3-digit. The first digit indicates a general status
- 1xx = informational message to be skipped (eg: 100,
- 2xx = OK, content is following (eg: 200,
- 3xx = OK, no content following (eg: 302,
- 4xx = error caused by the client (eg: 401, 403,
- 5xx = error caused by the server (eg: 500, 502,
Please refer to RFC2616 for the detailed meaning of all such codes.
"reason" field is just a hint, but is not parsed by clients. Anything can
found there, but it's a common practice to respect the
messages. It can be composed of one or multiple words, such as "OK",
or "Authentication
Haproxy may emit the following status codes by itself
Code When /
200 access to stats page, and when replying to monitoring
301 when performing a redirection, depending on the configured
302 when performing a redirection, depending on the configured
303 when performing a redirection, depending on the configured
400 for an invalid or too large
401 when an authentication is required to perform the action
accessing the stats
403 when a request is forbidden by a "block" ACL or "reqdeny"
408 when the request timeout strikes before the request is
500 when haproxy encounters an unrecoverable internal error, such as
memory allocation failure, which should never
502 when the server returns an empty, invalid or incomplete response,
when an "rspdeny" filter blocks the
503 when no server was available to handle the request, or in response
monitoring requests which match the "monitor fail"
504 when the response timeout strikes before the server
The error 4xx and 5xx codes above may be customized (see "errorloc" in
4.
1.3.2. The response
---------------------------
---------------------------
Response headers work exactly like request headers, and as such, HAProxy
the same parsing function for both. Please refer to paragraph 1.2.2 for
details.
details.
2. Configuring
----------------------
----------------------
2.1. Configuration file
------------------------------
------------------------------
HAProxy's configuration process involves 3 major sources of parameters
- the arguments from the command-line, which always take
- the "global" section, which sets process-wide
- the proxies sections which can take form of "defaults",
"frontend" and
The configuration file syntax consists in lines beginning with a
referenced in this manual, optionally followed by one or several
delimited by spaces. If spaces have to be entered in strings, then they must
preceded by a backslash ('\') to be escaped. Backslashes also have to
escaped by doubling
2.2. Time
----------------
----------------
Some parameters involve values representing time, such as timeouts.
values are generally expressed in milliseconds (unless explicitly
otherwise) but may be expressed in any other unit by suffixing the unit to
numeric value. It is important to consider this because it will not be
for every keyword. Supported units are
- us : microseconds. 1 microsecond = 1/1000000
- ms : milliseconds. 1 millisecond = 1/1000 second. This is the
- s : seconds. 1s =
- m : minutes. 1m = 60s =
- h : hours. 1h = 60m = 3600s =
- d : days. 1d = 24h = 1440m = 86400s =
2.3.
-------------
-------------
# Simple configuration for an HTTP proxy listening on port 80 on
# interfaces and forwarding requests to a single backend "servers" with
# single server "server1" listening on 127.0.0.
global
daemon
global
daemon
maxconn
defaults
defaults
mode
timeout connect
timeout client
timeout server
frontend
bind
default_backend
backend
server server1 127.0.0.1:8000 maxconn
# The same configuration defined with a single listen block. Shorter
# less expressive, especially in HTTP
global
daemon
global
daemon
maxconn
defaults
defaults
mode
timeout connect
timeout client
timeout server
listen
bind
server server1 127.0.0.1:8000 maxconn
Assuming haproxy is in $PATH, test these configurations in a shell
$ sudo haproxy -f configuration.conf
3. Global
--------------------
--------------------
Parameters in the "global" section are process-wide and often OS-specific.
are generally set once for all and do not need being changed once correct.
of them have command-line
The following keywords are supported in the "global" section
* Process management and
-
-
-
-
-
-
-
-
-
-
-
-
-
-
* Performance
-
-
-
-
-
-
-
-
- tune.
- tune.
- tune.
- tune.
- tune.
- tune.rcvbuf.
- tune.rcvbuf.
- tune.sndbuf.
- tune.sndbuf.
*
-
-
3.1. Process management and
------------------------------------
------------------------------------
chroot <jail
Changes current directory to <jail dir> and performs a chroot() there
dropping privileges. This increases the security level in case an
vulnerability would be exploited, since it would make it very hard for
attacker to exploit the system. This only works when the process is
with superuser privileges. It is important to ensure that <jail_dir> is
empty and unwritable to
daemon
daemon
Makes the process fork into background. This is the recommended mode
operation. It is equivalent to the command line "-D" argument. It can
disabled by the command line "-db"
gid
Changes the process' group ID to <number>. It is recommended that the
ID is dedicated to HAProxy or to a small set of similar daemons. HAProxy
be started with a user belonging to this group, or with superuser
See also "group" and
group <group
Similar to "gid" but uses the GID of group name <group name> from
See also "gid" and
log <address> <facility> [max level [min
Adds a global syslog server. Up to two global servers can be defined.
will receive logs for startups and exits, as well as all logs from
configured with "log
<address> can be one
- An IPv4 address optionally followed by a colon and a UDP port.
no port is specified, 514 is used by default (the standard
port).
port).
- A filesystem path to a UNIX domain socket, keeping in
considerations for chroot (be sure the path is accessible
the chroot) and uid/gid (be sure the path is
writeable).
writeable).
<facility> must be one of the 24 standard syslog facilities
kern user mail daemon auth syslog lpr
uucp cron auth2 ftp ntp audit alert
local0 local1 local2 local3 local4 local5 local6
An optional level can be specified to filter outgoing messages. By
all messages are sent. If a maximum level is specified, only messages with
severity at least as important as this level will be sent. An optional
level can be specified. If it is set, logs emitted with a more severe
than this one will be capped to this level. This is used to avoid
"emerg" messages on all terminals on some default syslog
Eight levels are known
emerg alert crit err warning notice info
log-send-hostname
Sets the hostname field in the syslog header. If optional "string"
is set the header is set to the string contents, otherwise uses the
of the system. Generally used if one is not relaying logs through
intermediate syslog server or for simply customizing the hostname printed
the
log-tag
Sets the tag field in the syslog header to this string. It defaults to
program name as launched from the command line, which usually is
Sometimes it can be useful to differentiate between multiple
running on the same
nbproc
Creates <number> processes when going daemon. This requires the
mode. By default, only one process is created, which is the recommended
of operation. For systems limited to small sets of file descriptors
process, it may be needed to fork multiple daemons. USING MULTIPLE
IS HARDER TO DEBUG AND IS REALLY DISCOURAGED. See also
pidfile
Writes pids of all daemons into file <pidfile>. This option is equivalent
the "-p" command line argument. The file must be accessible to the
starting the process. See also
stats socket <path> [{uid | user} <uid>] [{gid | group} <gid>] [mode
[level
Creates a UNIX socket in stream mode at location <path>. Any
existing socket will be backed up then replaced. Connections to this
will return various statistics outputs and even allow some commands to
issued. Please consult section 9.2 "Unix Socket commands" for more
An optional "level" parameter can be specified to restrict the nature
the commands that can be issued on the socket
- "user" is the least privileged level ; only non-sensitive stats can
read, and no change is allowed. It would make sense on systems where
is not easy to restrict access to the
- "operator" is the default level and fits most common uses. All data
be read, and only non-sensitive changes are permitted (eg: clear
counters).
counters).
- "admin" should be used with care, as everything is permitted (eg:
all
On platforms which support it, it is possible to restrict access to
socket by specifying numerical IDs after "uid" and "gid", or valid user
group names after the "user" and "group" keywords. It is also possible
restrict permissions on the socket by passing an octal value after the
keyword (same syntax as chmod). Depending on the platform, the permissions
the socket will be inherited from the directory which hosts it, or from
user the process is started
stats timeout <timeout, in
The default timeout on the stats socket is set to 10 seconds. It is
to change this value with "stats timeout". The value must be passed
milliseconds, or be suffixed by a time unit among { us, ms, s, m, h, d
stats maxconn
By default, the stats socket is limited to 10 concurrent connections. It
possible to change this value with "stats
uid
Changes the process' user ID to <number>. It is recommended that the user
is dedicated to HAProxy or to a small set of similar daemons. HAProxy
be started with superuser privileges in order to be able to switch to
one. See also "gid" and
ulimit-n
Sets the maximum number of per-process file-descriptors to <number>.
default, it is automatically computed, so it is recommended not to use
option.
option.
user <user
Similar to "uid" but uses the UID of user name <user name> from
See also "uid" and
node
Only letters, digits, hyphen and underscore are allowed, like in DNS
This statement is useful in HA configurations where two or more processes
servers share the same IP address. By setting a different node-name on
nodes, it becomes easy to immediately spot what server is handling
traffic.
traffic.
description
Add a text that describes the
Please note that it is required to escape certain characters (# for
and this text is inserted into a html page so you should avoid
"<" and ">"
3.2. Performance
-----------------------
-----------------------
maxconn
Sets the maximum per-process number of concurrent connections to <number>.
is equivalent to the command-line argument "-n". Proxies will stop
connections when this limit is reached. The "ulimit-n" parameter
automatically adjusted according to this value. See also
maxpipes
Sets the maximum per-process number of pipes to <number>. Currently,
are only used by kernel-based tcp splicing. Since a pipe contains two
descriptors, the "ulimit-n" value will be increased accordingly. The
value is maxconn/4, which seems to be more than enough for most heavy
The splice code dynamically allocates and releases pipes, and can fall
to standard copy, so setting this value too low may only impact
noepoll
noepoll
Disables the use of the "epoll" event polling system on Linux. It
equivalent to the command-line argument "-de". The next polling
used will generally be "poll". See also "nosepoll", and
nokqueue
nokqueue
Disables the use of the "kqueue" event polling system on BSD. It
equivalent to the command-line argument "-dk". The next polling
used will generally be "poll". See also
nopoll
nopoll
Disables the use of the "poll" event polling system. It is equivalent to
command-line argument "-dp". The next polling system used will be
It should never be needed to disable "poll" since it's available on
platforms supported by HAProxy. See also "nosepoll", and "nopoll"
"nokqueue".
nosepoll
"nokqueue".
nosepoll
Disables the use of the "speculative epoll" event polling system on Linux.
is equivalent to the command-line argument "-ds". The next polling
used will generally be "epoll". See also "nosepoll", and
nosplice
nosplice
Disables the use of kernel tcp splicing between sockets on Linux. It
equivalent to the command line argument "-dS". Data will then be
using conventional and more portable recv/send calls. Kernel tcp splicing
limited to some very recent instances of kernel 2.6. Most versions
2.6.25 and 2.6.28 are buggy and will forward corrupted data, so they must
be used. This option makes it easier to globally disable kernel splicing
case of doubt. See also "option splice-auto", "option splice-request"
"option
spread-checks <0..50, in
Sometimes it is desirable to avoid sending health checks to servers at
intervals, for instance when many logical servers are located on the
physical server. With the help of this parameter, it becomes possible to
some randomness in the check interval between 0 and +/- 50%. A value
2 and 5 seems to show good results. The default value remains at
tune.bufsize
Sets the buffer size to this size (in bytes). Lower values allow
sessions to coexist in the same amount of RAM, and higher values allow
applications with very large cookies to work. The default value is 16384
can be changed at build time. It is strongly recommended not to change
from the default value, as very low values will break some services such
statistics, and values larger than default size will increase memory
possibly causing the system to run out of memory. At least the global
parameter should be decreased by the same factor as this one is
tune.chksize
Sets the check buffer size to this size (in bytes). Higher values may
find string or regex patterns in very large pages, though doing so may
more memory and CPU usage. The default value is 16384 and can be changed
build time. It is not recommended to change this value, but to use
checks whenever
tune.maxaccept
Sets the maximum number of consecutive accepts that a process may perform
a single wake up. High values give higher priority to high connection
while lower values give higher priority to already established
This value is limited to 100 by default in single process mode. However,
multi-process mode (nbproc > 1), it defaults to 8 so that when one
wakes up, it does not take all incoming connections for itself and leaves
part of them to other processes. Setting this value to -1 completely
the limitation. It should normally not be needed to tweak this
tune.maxpollevents
Sets the maximum amount of events that can be processed at once in a call
the polling system. The default value is adapted to the operating system.
has been noticed that reducing it below 200 tends to slightly
latency at the expense of network bandwidth, and increasing it above
tends to trade latency for slightly increased
tune.maxrewrite
Sets the reserved buffer space to this size in bytes. The reserved space
used for header rewriting or appending. The first reads on sockets will
fill more than bufsize-maxrewrite. Historically it has defaulted to half
bufsize, though that does not make much sense since there are rarely
numbers of headers to add. Setting it too high prevents processing of
requests or responses. Setting it too low prevents addition of new
to already large requests or to POST requests. It is generally wise to set
to about 1024. It is automatically readjusted to half of bufsize if it
larger than that. This means you don't have to worry about it when
bufsize.
bufsize.
tune.rcvbuf.client
tune.rcvbuf.server
Forces the kernel socket receive buffer size on the client or the server
to the specified value in bytes. This value applies to all TCP/HTTP
and backends. It should normally never be set, and the default size (0)
the kernel autotune this value depending on the amount of available
However it can sometimes help to set it to very low values (eg: 4096)
order to save kernel memory by preventing it from buffering too large
of received data. Lower values will significantly increase CPU usage
tune.sndbuf.client
tune.sndbuf.server
Forces the kernel socket send buffer size on the client or the server side
the specified value in bytes. This value applies to all TCP/HTTP
and backends. It should normally never be set, and the default size (0)
the kernel autotune this value depending on the amount of available
However it can sometimes help to set it to very low values (eg: 4096)
order to save kernel memory by preventing it from buffering too large
of received data. Lower values will significantly increase CPU usage
Another use case is to prevent write timeouts with extremely slow clients
to the kernel waiting for a large part of the buffer to be read
notifying haproxy
3.3.
--------------
debug
--------------
debug
Enables debug mode which dumps to stdout all exchanges, and disables
into background. It is the equivalent of the command-line argument "-d".
should never be used in a production configuration since it may prevent
system
quiet
quiet
Do not display any message during startup. It is equivalent to the
line argument
3.4.
--------------
--------------
It is possible to control access to frontend/backend/listen sections or
http stats by allowing only authenticated and authorized users. To do
it is required to create at least one userlist and to define
userlist
Creates new userlist with name <listname>. Many independent userlists can
used to store authentication & authorization data for independent
group <groupname> [users <user>,<user>,(...
Adds group <groupname> to the current userlist. It is also possible
attach users to this group by using a comma separated list of
proceeded by "users"
user <username> [password|insecure-password
[groups <group>,<group>,(...
Adds user <username> to the current userlist. Both secure (encrypted)
insecure (unencrypted) passwords can be used. Encrypted passwords
evaluated using the crypt(3) function so depending of the
capabilities, different algorithms are supported. For example modern
based Linux system supports MD5, SHA-256, SHA-512 and of course
DES-based method of crypting
Example:
Example:
userlist
group G1 users
group G2 users
user tiger password $6$k6y3o.eP$JlKBx9za9667qe4(...)xHSwRv6J.
user scott insecure-password
user xdb insecure-password
userlist
group
group
user tiger password $6$k6y3o.eP$JlKBx(...)xHSwRv6J.C0/D7cV91 groups
user scott insecure-password elgato groups
user xdb insecure-password hello groups
Please note that both lists are functionally
4.
----------
----------
Proxy configuration can be located in a set of sections
- defaults
- frontend
- backend
- listen
A "defaults" section sets default parameters for all other sections
its declaration. Those default parameters are reset by the next
section. See below for the list of parameters which can be set in a
section. The name is optional but its use is encouraged for better
A "frontend" section describes a set of listening sockets accepting
connections.
connections.
A "backend" section describes a set of servers to which the proxy will
to forward incoming
A "listen" section defines a complete proxy with its frontend and
parts combined in one section. It is generally useful for TCP-only
All proxy names must be formed from upper and lower case letters,
'-' (dash), '_' (underscore) , '.' (dot) and ':' (colon). ACL names
case-sensitive, which means that "www" and "WWW" are two different
Historically, all proxy names could overlap, it just caused troubles in
logs. Since the introduction of content switching, it is mandatory that
proxies with overlapping capabilities (frontend/backend) have different
However, it is still permitted that a frontend and a backend share the
name, as this configuration seems to be commonly
Right now, two major proxy modes are supported : "tcp", also known as layer
and "http", also known as layer 7. In layer 4 mode, HAProxy simply
bidirectional traffic between two sides. In layer 7 mode, HAProxy analyzes
protocol, and can interact with it by allowing, blocking, switching,
modifying, or removing arbitrary contents in requests or responses, based
arbitrary
4.1. Proxy keywords
--------------------------
--------------------------
The following list of keywords is supported. Most of them may only be used in
limited set of section types. Some of them are marked as "deprecated"
they are inherited from an old syntax which may be confusing or
limited, and there are new recommended keywords to replace them.
marked with "(*)" can be optionally inverted using the "no" prefix, eg.
option contstats". This makes sense when the option has been enabled by
and must be disabled for a specific instance. Such options may also be
with "default" in order to restore default settings regardless of what has
specified in a previous "defaults"
keyword defaults frontend listen
------------------------------------+----------+----------+---------+---------
------------------------------------+----------+----------+---------+---------
acl - X X
appsession - - X
backlog X X X
balance X - X
bind - X X
bind-process X X X
block - X X
capture cookie - X X
capture request header - X X
capture response header - X X
clitimeout (deprecated) X X X
contimeout (deprecated) X - X
cookie X - X
default-server X - X
default_backend X X X
description - X X
disabled X X X
dispatch - - X
enabled X X X
errorfile X X X
errorloc X X X
errorloc302 X X X
-- keyword -------------------------- defaults - frontend - listen -- backend
errorloc303 X X X
force-persist - X X
fullconn X - X
grace X X X
hash-type X - X
http-check disable-on-404 X - X
http-check expect - - X
http-check send-state X - X
http-request - X X
id - X X
ignore-persist - X X
log X X X
maxconn X X X
mode X X X
monitor fail - X X
monitor-net X X X
monitor-uri X X X
option abortonclose (*) X - X
option accept-invalid-http-request (*) X X X
option accept-invalid-http-response (*) X - X
option allbackups (*) X - X
option checkcache (*) X - X
option clitcpka (*) X X X
option contstats (*) X X X
option dontlog-normal (*) X X X
option dontlognull (*) X X X
option forceclose (*) X X X
-- keyword -------------------------- defaults - frontend - listen -- backend
option forwardfor X X X
option http-no-delay (*) X X X
option http-pretend-keepalive (*) X X X
option http-server-close (*) X X X
option http-use-proxy-header (*) X X X
option httpchk X - X
option httpclose (*) X X X
option httplog X X X
option http_proxy (*) X X X
option independant-streams (*) X X X
option ldap-check X - X
option log-health-checks (*) X - X
option log-separate-errors (*) X X X
option logasap (*) X X X
option mysql-check X - X
option nolinger (*) X X X
option originalto X X X
option persist (*) X - X
option redispatch (*) X - X
option smtpchk X - X
option socket-stats (*) X X X
option splice-auto (*) X X X
option splice-request (*) X X X
option splice-response (*) X X X
option srvtcpka (*) X - X
option ssl-hello-chk X - X
-- keyword -------------------------- defaults - frontend - listen -- backend
option tcp-smart-accept (*) X X X
option tcp-smart-connect (*) X - X
option tcpka X X X
option tcplog X X X
option transparent (*) X - X
persist rdp-cookie X - X
rate-limit sessions X X X
redirect - X X
redisp (deprecated) X - X
redispatch (deprecated) X - X
reqadd - X X
reqallow - X X
reqdel - X X
reqdeny - X X
reqiallow - X X
reqidel - X X
reqideny - X X
reqipass - X X
reqirep - X X
reqisetbe - X X
reqitarpit - X X
reqpass - X X
reqrep - X X
-- keyword -------------------------- defaults - frontend - listen -- backend
reqsetbe - X X
reqtarpit - X X
retries X - X
rspadd - X X
rspdel - X X
rspdeny - X X
rspidel - X X
rspideny - X X
rspirep - X X
rsprep - X X
server - - X
source X - X
srvtimeout (deprecated) X - X
stats admin - - X
stats auth X - X
stats enable X - X
stats hide-version X - X
stats http-request - - X
stats realm X - X
stats refresh X - X
stats scope X - X
stats show-desc X - X
stats show-legends X - X
stats show-node X - X
stats uri X - X
-- keyword -------------------------- defaults - frontend - listen -- backend
stick match - - X
stick on - - X
stick store-request - - X
stick-table - - X
tcp-request content accept - X X
tcp-request content reject - X X
tcp-request inspect-delay - X X
timeout check X - X
timeout client X X X
timeout clitimeout (deprecated) X X X
timeout connect X - X
timeout contimeout (deprecated) X - X
timeout http-keep-alive X X X
timeout http-request X X X
timeout queue X - X
timeout server X - X
timeout srvtimeout (deprecated) X - X
timeout tarpit X X X
transparent (deprecated) X - X
use_backend - X X
------------------------------------+----------+----------+---------+---------
------------------------------------+----------+----------+---------+---------
keyword defaults frontend listen
4.2. Alphabetically sorted keywords
---------------------------------------------
---------------------------------------------
This section provides a description of each keyword and its
acl <aclname> <criterion> [flags] [operator]
Declare or complete an access
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Example:
Example:
acl invalid_src src 0.0.0.0/7 224.0.0.
acl invalid_src src_port
acl local_dst hdr(host) -i
See section 7 about ACL
appsession <cookie> len <length> timeout
[request-learn] [prefix] [mode
Define session stickiness on an existing application
May be used in sections : defaults | frontend | listen |
no | no | yes |
Arguments
<cookie> this is the name of the cookie used by the application and
HAProxy will have to learn for each new
<length> this is the max number of characters that will be memorized
checked in each cookie
<holdtime> this is the time after which the cookie will be removed
memory if unused. If no unit is specified, this time is
milliseconds.
request-learn
milliseconds.
request-learn
If this option is specified, then haproxy will be able to
the cookie found in the request in case the server does
specify any in response. This is typically what happens
PHPSESSID cookies, or when haproxy's session expires
the application's session and the correct server is
It is recommended to specify this option to improve
prefix When this option is specified, haproxy will match on the
prefix (or URL parameter prefix). The appsession value is
data following this
Example
appsession ASPSESSIONID len 64 timeout 3h
This will match the cookie
the appsession value will be
mode This option allows to change the URL parser
2 modes are currently supported
- path-parameters
The parser looks for the appsession in the path
part (each parameter is separated by a semi-colon), which
convenient for JSESSIONID for
This is the default mode if the option is not
- query-string
In this mode, the parser will look for the appsession in
query
When an application cookie is defined in a backend, HAProxy will check
the server sets such a cookie, and will store its value in a table,
associate it with the server's identifier. Up to <length> characters
the value will be retained. On each connection, haproxy will look for
cookie both in the "Cookie:" headers, and as a URL parameter (depending
the mode used). If a known value is found, the client will be directed to
server associated with this value. Otherwise, the load balancing algorithm
applied. Cookies are automatically removed from memory when they have
unused for a duration longer than
The definition of an application cookie is limited to one per
Note : Consider not using this feature in multi-process mode (nbproc >
unless you know what you do : memory is not shared between
processes, which can result in random
Example
appsession JSESSIONID len 52 timeout
See also : "cookie", "capture cookie", "balance", "stick",
"ignore-persist", "nbproc" and
backlog
Give hints to the system about the approximate listen backlog desired
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<conns> is the number of pending connections. Depending on the
system, it may represent the number of already
connections, of non-acknowledged ones, or
In order to protect against SYN flood attacks, one solution is to
the system's SYN backlog size. Depending on the system, sometimes it is
tunable via a system parameter, sometimes it is not adjustable at all,
sometimes the system relies on hints given by the application at the time
the listen() syscall. By default, HAProxy passes the frontend's maxconn
to the listen() syscall. On systems which can make use of this value, it
sometimes be useful to be able to specify a different value, hence
backlog
On Linux 2.4, the parameter is ignored by the system. On Linux 2.6, it
used as a hint and the system accepts up to the smallest greater power
two, and never more than some limits (usually
See also : "maxconn" and the target operating system's tuning
balance <algorithm> [ <arguments>
balance url_param <param> [check_post
Define the load balancing algorithm to be used in a
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<algorithm> is the algorithm used to select a server when doing
balancing. This only applies when no persistence
is available, or when a connection is redispatched to
server. <algorithm> may be one of the following
roundrobin Each server is used in turns, according to their
This is the smoothest and fairest algorithm when the
processing time remains equally distributed. This
is dynamic, which means that server weights may be
on the fly for slow starts for instance. It is limited
design to 4128 active servers per backend. Note that in
large farms, when a server becomes up after having been
for a very short time, it may sometimes take a few
requests for it to be re-integrated into the farm and
receiving traffic. This is normal, though very rare. It
indicated here in case you would have the chance to
it, so that you don't
static-rr Each server is used in turns, according to their
This algorithm is as similar to roundrobin except that it
static, which means that changing a server's weight on
fly will have no effect. On the other hand, it has no
limitation on the number of servers, and when a server
up, it is always immediately reintroduced into the farm,
the full map is recomputed. It also uses slightly less CPU
run (around
leastconn The server with the lowest number of connections receives
connection. Round-robin is performed within groups of
of the same load to ensure that all servers will be used.
of this algorithm is recommended where very long sessions
expected, such as LDAP, SQL, TSE, etc... but is not very
suited for protocols using short sessions such as HTTP.
algorithm is dynamic, which means that server weights may
adjusted on the fly for slow starts for
source The source IP address is hashed and divided by the
weight of the running servers to designate which server
receive the request. This ensures that the same client
address will always reach the same server as long as
server goes down or up. If the hash result changes due to
number of running servers changing, many clients will
directed to a different server. This algorithm is
used in TCP mode where no cookie may be inserted. It may
be used on the Internet to provide a best-effort
to clients which refuse session cookies. This algorithm
static by default, which means that changing a
weight on the fly will have no effect, but this can
changed using
uri This algorithm hashes either the left part of the URI
the question mark) or the whole URI (if the "whole"
is present) and divides the hash value by the total weight
the running servers. The result designates which server
receive the request. This ensures that the same URI
always be directed to the same server as long as no
goes up or down. This is used with proxy caches
anti-virus proxies in order to maximize the cache hit
Note that this algorithm may only be used in an HTTP
This algorithm is static by default, which means
changing a server's weight on the fly will have no
but this can be changed using
This algorithm supports two optional parameters "len"
"depth", both followed by a positive integer number.
options may be helpful when it is needed to balance
based on the beginning of the URI only. The "len"
indicates that the algorithm should only consider that
characters at the beginning of the URI to compute the
Note that having "len" set to 1 rarely makes sense since
URIs start with a leading
The "depth" parameter indicates the maximum directory
to be used to compute the hash. One level is counted for
slash in the request. If both parameters are specified,
evaluation stops when either is
url_param The URL parameter specified in argument will be looked up
the query string of each HTTP GET
If the modifier "check_post" is used, then an HTTP
request entity will be searched for the parameter
when it is not found in a query string after a question
('?') in the URL. Optionally, specify a number of octets
wait for before attempting to search the message body. If
entity can not be searched, then round robin is used for
request. For instance, if your clients always send the
parameter in the first 128 bytes, then specify that.
default is 48. The entity data will not be scanned until
required number of octets have arrived at the gateway,
is the minimum of: (default/max_wait, Content-Length or
chunk length). If Content-Length is missing or zero, it
not need to wait for more data than the client promised
send. When Content-Length is present and larger
<max_wait>, then waiting is limited to <max_wait> and it
assumed that this will be enough data to search for
presence of the parameter. In the unlikely event
Transfer-Encoding: chunked is used, only the first chunk
scanned. Parameter values separated by a chunk boundary,
be randomly balanced if at
If the parameter is found followed by an equal sign ('=')
a value, then the value is hashed and divided by the
weight of the running servers. The result designates
server will receive the
This is used to track user identifiers in requests and
that a same user ID will always be sent to the same server
long as no server goes up or down. If no value is found or
the parameter is not found, then a round robin algorithm
applied. Note that this algorithm may only be used in an
backend. This algorithm is static by default, which
that changing a server's weight on the fly will have
effect, but this can be changed using
hdr(<name>) The HTTP header <name> will be looked up in each HTTP
Just as with the equivalent ACL 'hdr()' function, the
name in parenthesis is not case sensitive. If the header
absent or if it does not contain any value, the
algorithm is applied
An optional 'use_domain_only' parameter is available,
reducing the hash algorithm to the main domain part with
specific headers such as 'Host'. For instance, in the
value "haproxy.1wt.eu", only "1wt" will be
This algorithm is static by default, which means
changing a server's weight on the fly will have no
but this can be changed using
rdp-cookie
rdp-cookie(name)
rdp-cookie
rdp-cookie(name)
The RDP cookie <name> (or "mstshash" if omitted) will
looked up and hashed for each incoming TCP request. Just
with the equivalent ACL 'req_rdp_cookie()' function, the
is not case-sensitive. This mechanism is useful as a
persistence mode, as it makes it possible to always send
same user (or the same session ID) to the same server. If
cookie is not found, the normal roundrobin algorithm
used
Note that for this to work, the frontend must ensure that
RDP cookie is already present in the request buffer. For
you must use 'tcp-request content accept' rule combined
a 'req_rdp_cookie_cnt'
This algorithm is static by default, which means
changing a server's weight on the fly will have no
but this can be changed using
<arguments> is an optional list of arguments which may be needed by
algorithms. Right now, only "url_param" and "uri" support
optional
balance uri [len <len>] [depth
balance url_param <param> [check_post
The load balancing algorithm of a backend is set to roundrobin when no
algorithm, mode nor option have been set. The algorithm may only be set
for each
Examples
balance
balance url_param
balance url_param session_id check_post
balance
balance
balance hdr(Host)
Note: the following caveats and limitations on using the
extension with "url_param" must be considered
- all POST requests are eligible for consideration, because there is no
to determine if the parameters will be found in the body or entity
may contain binary data. Therefore another method may be required
restrict consideration of POST requests that have no URL parameters
the body. (see acl reqideny
- using a <max_wait> value larger than the request buffer size does
make sense and is useless. The buffer size is set at build time,
defaults to 16
- Content-Encoding is not supported, the parameter search will
fail; and load balancing will fall back to Round
- Expect: 100-continue is not supported, load balancing will fall back
Round
- Transfer-Encoding (RFC2616 3.6.1) is only supported in the first
If the entire parameter value is not present in the first chunk,
selection of server is undefined (actually, defined by how
actually appeared in the first
- This feature does not support generation of a 100, 411 or 501
- In some cases, requesting "check_post" MAY attempt to scan the
contents of a message body. Scanning normally terminates when
white space or control characters are found, indicating the end of
might be a URL parameter list. This is probably not a concern with
type message
See also : "dispatch", "cookie", "appsession", "transparent", "hash-type"
"http_proxy".
"http_proxy".
bind [<address>]:<port_range> [, ...
bind [<address>]:<port_range> [, ...] interface
bind [<address>]:<port_range> [, ...] mss
bind [<address>]:<port_range> [, ...]
bind [<address>]:<port_range> [, ...] id
bind [<address>]:<port_range> [, ...] name
bind [<address>]:<port_range> [, ...]
Define one or several listening addresses and/or ports in a
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<address> is optional and can be a host name, an IPv4 address, an
address, or '*'. It designates the address the frontend
listen on. If unset, all IPv4 addresses of the system will
listened on. The same will apply for '*' or the
special address "0.0.0.
<port_range> is either a unique TCP port, or a port range for which
proxy will accept connections for the IP address
above. The port is mandatory. Note that in the case of
IPv6 address, the port is always the number after the
colon (':'). A range can either be
- a numerical port (ex:
- a dash-delimited ports range explicitly stating the
and upper bounds (ex: '2000-2100') which are included
the
Particular care must be taken against port ranges,
every <address:port> couple consumes one socket (= a
descriptor), so it's easy to consume lots of
with a simple range, and to run out of sockets. Also,
<address:port> couple must be used only once among
instances running on a same system. Please note that
to ports lower than 1024 generally require
privileges to start the program, which are independant
the 'uid'
<interface> is an optional physical interface name. This is
only supported on Linux. The interface must be a
interface, not an aliased interface. When specified,
addresses on the same line will only be accepted if
incoming packet physically come through the
interface. It is also possible to bind multiple frontends
the same address if they are bound to different
Note that binding to a physical interface requires
privileges.
privileges.
<maxseg> is an optional TCP Maximum Segment Size (MSS) value to
advertised on incoming connections. This can be used to
a lower MSS for certain specific ports, for instance
connections passing through a VPN. Note that this relies on
kernel feature which is theorically supported under Linux
was buggy in all versions prior to 2.6.28. It may or may
work on other operating systems. The commonly
value on Ethernet networks is 1460 = 1500(MTU) -
<id> is a persistent value for socket ID. Must be positive
unique in the proxy. An unused value will automatically
assigned if unset. Can only be used when defining only
single
<name> is an optional name provided for
transparent is an optional keyword which is supported only on
Linux kernels. It indicates that the addresses will be
even if they do not belong to the local machine. Any
targeting any of these addresses will be caught just as
the address was locally configured. This normally
that IP forwarding is enabled. Caution! do not use this
the default address '*', as it would redirect any traffic
the specified port. This keyword is available only
HAProxy is built with
defer-accept is an optional keyword which is supported only on
Linux kernels. It states that a connection will only
accepted once some data arrive on it, or at worst after
first retransmit. This should be used only on protocols
which the client talks first (eg: HTTP). It can
improve performance by ensuring that most of the request
already available when the connection is accepted. On
other hand, it will not be able to detect connections
don't talk. It is important to note that this option
broken in all kernels up to 2.6.31, as the connection
never accepted until the client talks. This can cause
with front firewalls which would see an
connection while the proxy will only see it in
It is possible to specify a list of address:port combinations delimited
commas. The frontend will then listen on all of these addresses. There is
fixed limit to the number of addresses and ports which can be listened on
a frontend, as well as there is no limit to the number of "bind"
in a
Example
listen
bind
bind 10.0.0.1:10080,10.0.0.
See also :
bind-process [ all | odd | even | <number 1-32>
Limit visibility of an instance to a certain set of processes
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
all All process will see this instance. This is the default.
may be used to override a default
odd This instance will be enabled on processes 1,3,5,...31.
option may be combined with other
even This instance will be enabled on processes 2,4,6,...32.
option may be combined with other numbers. Do not use
with less than 2 processes otherwise some instances might
missing from all
number The instance will be enabled on this process number,
1 and 32. You must be careful not to reference a
number greater than the configured global.nbproc,
some instances might be missing from all
This keyword limits binding of certain instances to certain processes.
is useful in order not to have too many processes listening to the
ports. For instance, on a dual-core machine, it might make sense to
'nbproc 2' in the global section, then distributes the listeners among
and 'even'
At the moment, it is not possible to reference more than 32 processes
this keyword, but this should be more than enough for most setups.
note that 'all' really means all processes and is not limited to the
32.
32.
If some backends are referenced by frontends bound to other processes,
backend automatically inherits the frontend's
Example
listen
bind 10.0.0.
bind-process
listen
bind 10.0.0.
bind-process
listen
bind 10.0.0.
bind-process 1 2 3
See also : "nbproc" in global
block { if | unless }
Block a layer 7 request if/unless a condition is
May be used in sections : defaults | frontend | listen |
no | yes | yes |
The HTTP request will be blocked very early in the layer 7
if/unless <condition> is matched. A 403 error will be returned if the
is blocked. The condition has to reference ACLs (see section 7). This
typically used to deny access to certain sensitive resources if
conditions are met or not met. There is no fixed limit to the number
"block" statements per
Example:
Example:
acl invalid_src src 0.0.0.0/7 224.0.0.
acl invalid_src src_port
acl local_dst hdr(host) -i
block if invalid_src ||
See section 7 about ACL
capture cookie <name> len
Capture and log a cookie in the request and in the
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<name> is the beginning of the name of the cookie to capture. In
to match the exact name, simply suffix the name with an
sign ('='). The full name will appear in the logs, which
useful with application servers which adjust both the cookie
and value (eg:
<length> is the maximum number of characters to report in the logs,
include the cookie name, the equal sign and the value, all in
standard "name=value" form. The string will be truncated on
right if it exceeds
Only the first cookie is captured. Both the "cookie" request headers and
"set-cookie" response headers are monitored. This is particularly useful
check for application bugs causing session crossing or stealing
users, because generally the user's cookies can only change on a login
When the cookie was not presented by the client, the associated log
will report "-". When a request does not cause a cookie to be assigned by
server, a "-" is reported in the response
The capture is performed in the frontend only because it is necessary
the log format does not change for a given frontend depending on
backends. This may change in the future. Note that there can be only
"capture cookie" statement in a frontend. The maximum capture length
configured in the sources by default to 64 characters. It is not possible
specify a capture in a "defaults"
Example:
Example:
capture cookie ASPSESSION len
See also : "capture request header", "capture response header" as well
section 8 about
capture request header <name> len
Capture and log the first occurrence of the specified request
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<name> is the name of the header to capture. The header names are
case-sensitive, but it is a common practice to write them as
appear in the requests, with the first letter of each word
upper case. The header name will not appear in the logs, only
value is reported, but the position in the logs is
<length> is the maximum number of characters to extract from the value
report in the logs. The string will be truncated on the right
it exceeds
Only the first value of the last occurrence of the header is captured.
value will be added to the logs between braces ('{}'). If multiple
are captured, they will be delimited by a vertical bar ('|') and will
in the same order they were declared in the configuration.
headers will be logged just as an empty string. Common uses for
header captures include the "Host" field in virtual hosting environments,
"Content-length" when uploads are supported, "User-agent" to
differentiate between real users and robots, and "X-Forwarded-For" in
environments to find where the request came
Note that when capturing headers such as "User-agent", some spaces may
logged, making the log analysis more difficult. Thus be careful about
you log if you know your log parser is not smart enough to rely on
braces.
braces.
There is no limit to the number of captured request headers, but each
is limited to 64 characters. In order to keep log format consistent for
same frontend, header captures can only be declared in a frontend. It is
possible to specify a capture in a "defaults"
Example:
Example:
capture request header Host len
capture request header X-Forwarded-For len
capture request header Referrer len
See also : "capture cookie", "capture response header" as well as section
about
capture response header <name> len
Capture and log the first occurrence of the specified response
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<name> is the name of the header to capture. The header names are
case-sensitive, but it is a common practice to write them as
appear in the response, with the first letter of each word
upper case. The header name will not appear in the logs, only
value is reported, but the position in the logs is
<length> is the maximum number of characters to extract from the value
report in the logs. The string will be truncated on the right
it exceeds
Only the first value of the last occurrence of the header is captured.
result will be added to the logs between braces ('{}') after the
request headers. If multiple headers are captured, they will be delimited
a vertical bar ('|') and will appear in the same order they were declared
the configuration. Non-existent headers will be logged just as an
string. Common uses for response header captures include the
header which indicates how many bytes are expected to be returned,
"Location" header to track
There is no limit to the number of captured response headers, but
capture is limited to 64 characters. In order to keep log format
for a same frontend, header captures can only be declared in a frontend.
is not possible to specify a capture in a "defaults"
Example:
Example:
capture response header Content-length len
capture response header Location len
See also : "capture cookie", "capture request header" as well as section
about
clitimeout <timeout>
Set the maximum inactivity time on the client
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<timeout> is the timeout value is specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
The inactivity timeout applies when the client is expected to acknowledge
send data. In HTTP mode, this timeout is particularly important to
during the first phase, when the client sends the request, and during
response while it is reading data sent by the server. The value is
in milliseconds by default, but can be in any other unit if the number
suffixed by the unit, as specified at the top of this document. In TCP
(and to a lesser extent, in HTTP mode), it is highly recommended that
client timeout remains equal to the server timeout in order to avoid
situations to debug. It is a good practice to cover one or several TCP
losses by specifying timeouts that are slightly above multiples of 3
(eg: 4 or 5
This parameter is specific to frontends, but can be specified once for all
"defaults" sections. This is in fact one of the easiest solutions not
forget about it. An unspecified timeout results in an infinite timeout,
is not recommended. Such a usage is accepted and works but reports a
during startup because it may results in accumulation of expired sessions
the system if the system's timeouts are not configured
This parameter is provided for compatibility but is currently
Please use "timeout client"
See also : "timeout client", "timeout http-request", "timeout server",
"srvtimeout".
"srvtimeout".
contimeout <timeout>
Set the maximum time to wait for a connection attempt to a server to
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<timeout> is the timeout value is specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
If the server is located on the same LAN as haproxy, the connection should
immediate (less than a few milliseconds). Anyway, it is a good practice
cover one or several TCP packet losses by specifying timeouts that
slightly above multiples of 3 seconds (eg: 4 or 5 seconds). By default,
connect timeout also presets the queue timeout to the same value if this
has not been specified. Historically, the contimeout was also used to set
tarpit timeout in a listen section, which is not possible in a pure
This parameter is specific to backends, but can be specified once for all
"defaults" sections. This is in fact one of the easiest solutions not
forget about it. An unspecified timeout results in an infinite timeout,
is not recommended. Such a usage is accepted and works but reports a
during startup because it may results in accumulation of failed sessions
the system if the system's timeouts are not configured
This parameter is provided for backwards compatibility but is
deprecated. Please use "timeout connect", "timeout queue" or "timeout
instead.
instead.
See also : "timeout connect", "timeout queue", "timeout
"timeout server",
cookie <name> [ rewrite | insert | prefix ] [ indirect ] [ nocache
[ postonly ] [ preserve ] [ httponly ] [ secure
[ domain <domain> ]* [ maxidle <idle> ] [ maxlife <life>
Enable cookie-based persistence in a
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<name> is the name of the cookie which will be monitored, modified
inserted in order to bring persistence. This cookie is sent
the client via a "Set-Cookie" header in the response, and
brought back by the client in a "Cookie" header in all
Special care should be taken to choose a name which does
conflict with any likely application cookie. Also, if the
backends are subject to be used by the same clients
HTTP/HTTPS), care should be taken to use different cookie
between all backends if persistence between them is not
rewrite This keyword indicates that the cookie will be provided by
server and that haproxy will have to modify its value to set
server's identifier in it. This mode is handy when the
of complex combinations of "Set-cookie" and
headers is left to the application. The application can
decide whether or not it is appropriate to emit a
cookie. Since all responses should be monitored, this mode
works in HTTP close mode. Unless the application behaviour
very complex and/or broken, it is advised not to start with
mode for new deployments. This keyword is incompatible
"insert" and
insert This keyword indicates that the persistence cookie will have
be inserted by haproxy in server responses if the client did
already have a cookie that would have permitted it to access
server. When used without the "preserve" option, if the
emits a cookie with the same name, it will be remove
processing. For this reason, this mode can be used to
existing configurations running in the "rewrite" mode. The
will only be a session cookie and will not be stored on
client's disk. By default, unless the "indirect" option is
the server will see the cookies emitted by the client. Due
caching effects, it is generally wise to add the "nocache"
"postonly" keywords (see below). The "insert" keyword is
compatible with "rewrite" and
prefix This keyword indicates that instead of relying on a
cookie for the persistence, an existing one will be
This may be needed in some specific environments where the
does not support more than one single cookie and the
already needs it. In this case, whenever the server sets a
named <name>, it will be prefixed with the server's
and a delimiter. The prefix will be removed from all
requests so that the server still finds the cookie it
Since all requests and responses are subject to being
this mode requires the HTTP close mode. The "prefix" keyword
not compatible with "rewrite" and
indirect When this option is specified, no cookie will be emitted to
client which already has a valid one for the server which
processed the request. If the server sets such a cookie
it will be removed, unless the "preserve" option is also set.
"insert" mode, this will additionally remove cookies from
requests transmitted to the server, making the
mechanism totally transparent from an application point of
nocache This option is recommended in conjunction with the insert
when there is a cache between the client and HAProxy, as
ensures that a cacheable response will be tagged non-cacheable
a cookie needs to be inserted. This is important because if
persistence cookies are added on a cacheable home page
instance, then all customers will then fetch the page from
outer cache and will all share the same persistence
leading to one server receiving much more traffic than
See also the "insert" and "postonly"
postonly This option ensures that cookie insertion will only be
on responses to POST requests. It is an alternative to
"nocache" option, because POST responses are not cacheable,
this ensures that the persistence cookie will never get
Since most sites do not need any sort of persistence before
first POST which generally is a login request, this is a
efficient method to optimize caching without risking to find
persistence cookie in the
See also the "insert" and "nocache"
preserve This option may only be used with "insert" and/or "indirect".
allows the server to emit the persistence cookie itself. In
case, if a cookie is found in the response, haproxy will leave
untouched. This is useful in order to end persistence after
logout request for instance. For this, the server just has
emit a cookie with an invalid value (eg: empty) or with a date
the past. By combining this mechanism with the
check option, it is possible to perform a completely
shutdown because users will definitely leave the server
they
httponly This option tells haproxy to add an "HttpOnly" cookie
when a cookie is inserted. This attribute is used so that
user agent doesn't share the cookie with non-HTTP
Please check RFC6265 for more information on this
secure This option tells haproxy to add a "Secure" cookie attribute
a cookie is inserted. This attribute is used so that a user
never emits this cookie over non-secure channels, which
that a cookie learned with this flag will be presented only
SSL/TLS connections. Please check RFC6265 for more information
this
domain This option allows to specify the domain at which a cookie
inserted. It requires exactly one parameter: a valid
name. If the domain begins with a dot, the browser is allowed
use it for any host ending with that name. It is also possible
specify several domain names by invoking this option
times. Some browsers might have small limits on the number
domains, so be careful when doing that. For the record,
10 domains to MSIE 6 or Firefox 2 works as
maxidle This option allows inserted cookies to be ignored after some
time. It only works with insert-mode cookies. When a cookie
sent to the client, the date this cookie was emitted is sent
Upon further presentations of this cookie, if the date is
than the delay indicated by the parameter (in seconds), it
be ignored. Otherwise, it will be refreshed if needed when
response is sent to the client. This is particularly useful
prevent users who never close their browsers from remaining
too long on the same server (eg: after a farm size change).
this option is set and a cookie has no date, it is
accepted, but gets refreshed in the response. This maintains
ability for admins to access their sites. Cookies that have
date in the future further than 24 hours are ignored. Doing
lets admins fix timezone issues without risking kicking users
the
maxlife This option allows inserted cookies to be ignored after some
time, whether they're in use or not. It only works with
mode cookies. When a cookie is first sent to the client, the
this cookie was emitted is sent too. Upon further
of this cookie, if the date is older than the delay indicated
the parameter (in seconds), it will be ignored. If the cookie
the request has no date, it is accepted and a date will be
Cookies that have a date in the future further than 24 hours
ignored. Doing so lets admins fix timezone issues without
kicking users off the site. Contrary to maxidle, this value
not refreshed, only the first visit date counts. Both maxidle
maxlife may be used at the time. This is particularly useful
prevent users who never close their browsers from remaining
too long on the same server (eg: after a farm size change).
is stronger than the maxidle method in that it forces
redispatch after some absolute
There can be only one persistence cookie per HTTP backend, and it can
declared in a defaults section. The value of the cookie will be the
indicated after the "cookie" keyword in a "server" statement. If no
is declared for a given server, the cookie is not
Examples
cookie JSESSIONID
cookie SRV insert indirect
cookie SRV insert postonly
cookie SRV insert indirect nocache maxidle 30m maxlife
See also : "appsession", "balance source", "capture cookie",
and
default-server
Change default options for a server in a
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments:
Arguments:
<param*> is a list of parameters for this server. The
keyword accepts an important number of options and has a
section dedicated to it. Please refer to section 5 for
details.
details.
Example
default-server inter 1000 weight
See also: "server" and section 5 about server
default_backend
Specify the backend to use when no "use_backend" rule has been
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<backend> is the name of the backend to
When doing content-switching between frontend and backends using
"use_backend" keyword, it is often useful to indicate which backend will
used when no rule has matched. It generally is the dynamic backend
will catch all undetermined
Example
use_backend dynamic if
use_backend static if url_css url_img
default_backend
See also : "use_backend", "reqsetbe",
disabled
disabled
Disable a proxy, frontend or
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
The "disabled" keyword is used to disable an instance, mainly in order
liberate a listening port or to temporarily disable a service. The
will still be created and its configuration will be checked, but it will
created in the "stopped" state and will appear as such in the statistics.
will not receive any traffic nor will it send any health-checks or logs.
is possible to disable many instances at once by adding the
keyword in a "defaults"
See also :
dispatch
Set a default server
May be used in sections : defaults | frontend | listen |
no | no | yes |
Arguments :
<address> is the IPv4 address of the default server. Alternatively,
resolvable hostname is supported, but this name will be
during
<ports> is a mandatory port specification. All connections will be
to this port, and it is not permitted to use port offsets as
possible with normal
The "dispatch" keyword designates a default server for use when no
server can take the connection. In the past it was used to forward
persistent connections to an auxiliary load balancer. Due to its
syntax, it has also been used for simple TCP relays. It is recommended not
use it for more clarity, and to use the "server" directive
See also :
enabled
enabled
Enable a proxy, frontend or
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
The "enabled" keyword is used to explicitly enable an instance, when
defaults has been set to "disabled". This is very rarely
See also :
errorfile <code>
Return a file contents instead of errors generated by
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<code> is the HTTP status code. Currently, HAProxy is capable
generating codes 200, 400, 403, 408, 500, 502, 503, and
<file> designates a file containing the full HTTP response. It
recommended to follow the common practice of appending ".http"
the filename so that people do not confuse the response with
error pages, and to use absolute paths, since files are
before any chroot is
It is important to understand that this keyword is not meant to
errors returned by the server, but errors detected and returned by
This is why the list of supported errors is limited to a small
Code 200 is emitted in response to requests matching a "monitor-uri"
The files are returned verbatim on the TCP socket. This allows any trick
as redirections to another URL or site, as well as tricks to clean
force enable or disable caching, etc... The package provides default
files returning the same contents as default
The files should not exceed the configured buffer size (BUFSIZE),
generally is 8 or 16 kB, otherwise they will be truncated. It is also
not to put any reference to local contents (eg: images) in order to
loops between the client and HAProxy when all servers are down, causing
error to be returned instead of an image. For better HTTP compliance, it
recommended that all header lines end with CR-LF and not LF
The files are read at the same time as the configuration and kept in
For this reason, the errors continue to be returned even when the process
chrooted, and no file change is considered while the process is running.
simple method for developing those files consists in associating them to
403 status code and interrogating a blocked
See also : "errorloc", "errorloc302",
Example
errorfile 400 /etc/haproxy/errorfiles/400badreq.
errorfile 403 /etc/haproxy/errorfiles/403forbid.
errorfile 503 /etc/haproxy/errorfiles/503sorry.
errorloc <code>
errorloc302 <code>
Return an HTTP redirection to a URL instead of errors generated by
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<code> is the HTTP status code. Currently, HAProxy is capable
generating codes 200, 400, 403, 408, 500, 502, 503, and
<url> it is the exact contents of the "Location" header. It may
either a relative URI to an error page hosted on the same
or an absolute URI designating an error page on another
Special care should be given to relative URIs to avoid
loops if the URI itself may generate the same error (eg:
It is important to understand that this keyword is not meant to
errors returned by the server, but errors detected and returned by
This is why the list of supported errors is limited to a small
Code 200 is emitted in response to requests matching a "monitor-uri"
Note that both keyword return the HTTP 302 status code, which tells
client to fetch the designated URL using the same HTTP method. This can
quite problematic in case of non-GET methods such as POST, because the
sent to the client might not be allowed for something other than GET.
workaround this problem, please use "errorloc303" which send the HTTP
status code, indicating to the client that the URL must be fetched with a
request.
request.
See also : "errorfile",
errorloc303 <code>
Return an HTTP redirection to a URL instead of errors generated by
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<code> is the HTTP status code. Currently, HAProxy is capable
generating codes 400, 403, 408, 500, 502, 503, and
<url> it is the exact contents of the "Location" header. It may
either a relative URI to an error page hosted on the same
or an absolute URI designating an error page on another
Special care should be given to relative URIs to avoid
loops if the URI itself may generate the same error (eg:
It is important to understand that this keyword is not meant to
errors returned by the server, but errors detected and returned by
This is why the list of supported errors is limited to a small
Code 200 is emitted in response to requests matching a "monitor-uri"
Note that both keyword return the HTTP 303 status code, which tells
client to fetch the designated URL using the same HTTP GET method.
solves the usual problems associated with "errorloc" and the 302 code. It
possible that some very old browsers designed before HTTP/1.1 do not
it, but no such problem has been reported till
See also : "errorfile", "errorloc",
force-persist { if | unless }
Declare a condition to force persistence on down
May be used in sections: defaults | frontend | listen |
no | yes | yes |
By default, requests are not dispatched to down servers. It is possible
force this using "option persist", but it is unconditional and
to a valid server if "option redispatch" is set. That leaves with very
possibilities to force some requests to reach a server which is
marked down for maintenance
The "force-persist" statement allows one to declare various
conditions which, when met, will cause a request to ignore the down status
a server and still try to connect to it. That makes it possible to start
server, still replying an error to the health checks, and run a
configured browser to test the service. Among the handy methods, one
use a specific source IP address, or a specific cookie. The cookie also
the advantage that it can easily be added/removed on the browser from a
page. Once the service is validated, it is then possible to open the
to the world by returning a valid response to health
The forced persistence is enabled when an "if" condition is met, or unless
"unless" condition is met. The final redispatch is always disabled when
is
See also : "option redispatch", "ignore-persist",
and section 7 about ACL
fullconn
Specify at what backend load the servers will reach their
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<conns> is the number of connections on the backend which will make
servers use the maximal number of
When a server has a "maxconn" parameter specified, it means that its
of concurrent connections will never go higher. Additionally, if it has
"minconn" parameter, it indicates a dynamic limit following the
load. The server will then always accept at least <minconn>
never more than <maxconn>, and the limit will be on the ramp between
values when the backend has less than <conns> concurrent connections.
makes it possible to limit the load on the servers during normal loads,
push it further for important loads without overloading the servers
exceptional
Example
# The servers will accept between 100 and 1000 concurrent connections
# and the maximum of 1000 will be reached when the backend reaches
#
backend
fullconn
server srv1 dyn1:80 minconn 100 maxconn
server srv2 dyn2:80 minconn 100 maxconn
See also : "maxconn",
grace
Maintain a proxy operational for some time after a soft
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<time> is the time (by default in milliseconds) for which the
will remain operational with the frontend sockets still
when a soft-stop is received via the SIGUSR1
This may be used to ensure that the services disappear in a certain
This was designed so that frontends which are dedicated to monitoring by
external equipment fail immediately while other ones remain up for the
needed by the equipment to detect the
Note that currently, there is very little benefit in using this
and it may in fact complicate the soft-reconfiguration process more
simplify
hash-type
Specify a method to use for mapping hashes to
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
map-based the hash table is a static array containing all alive
The hashes will be very smooth, will consider weights, but
be static in that weight changes while a server is up will
ignored. This means that there will be no slow start.
since a server is selected by its position in the array,
mappings are changed when the server count changes. This
that when a server goes up or down, or when a server is
to a farm, most connections will be redistributed to
servers. This can be inconvenient with caches for
consistent the hash table is a tree filled with many occurrences of
server. The hash key is looked up in the tree and the
server is chosen. This hash is dynamic, it supports
weights while the servers are up, so it is compatible with
slow start feature. It has the advantage that when a
goes up or down, only its associations are moved. When a
is added to the farm, only a few part of the mappings
redistributed, making it an ideal algorithm for
However, due to its principle, the algorithm will never be
smooth and it may sometimes be necessary to adjust a
weight or its ID to get a more balanced distribution. In
to get the same distribution on multiple load balancers, it
important that all servers have the same
The default hash type is "map-based" and is recommended for most
See also : "balance",
http-check
Enable a maintenance mode upon HTTP/404 response to
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
When this option is set, a server which returns an HTTP code 404 will
excluded from further load-balancing, but will still receive
connections. This provides a very convenient method for Web
to perform a graceful shutdown of their servers. It is also important to
that a server which is detected as failed while it was in this mode will
generate an alert, just a notice. If the server responds 2xx or 3xx again,
will immediately be reinserted into the farm. The status on the stats
reports "NOLB" for a server in this mode. It is important to note that
option only works in conjunction with the "httpchk" option. If this
is used with "http-check expect", then it has precedence over it so that
responses will still be considered as
See also : "option httpchk", "http-check
http-check expect [!] <match>
Make HTTP health checks consider reponse contents or specific status
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<match> is a keyword indicating how to look for a specific pattern in
response. The keyword may be one of "status",
"string", or "rstring". The keyword may be preceeded by
exclamation mark ("!") to negate the match. Spaces are
between the exclamation mark and the keyword. See below for
details on the supported
<pattern> is the pattern to look for. It may be a string or a
expression. If the pattern contains spaces, they must be
with the usual backslash
By default, "option httpchk" considers that response statuses 2xx and
are valid, and that others are invalid. When "http-check expect" is
it defines what is considered valid or invalid. Only one
statement is supported in a backend. If a server fails to respond or
out, the check obviously fails. The available matches are
status <string> : test the exact string match for the HTTP status
A health check respose will be considered valid if
response's status code is exactly this string. If
"status" keyword is prefixed with "!", then the
will be considered invalid if the status code
rstatus <regex> : test a regular expression for the HTTP status
A health check respose will be considered valid if
response's status code matches the expression. If
"rstatus" keyword is prefixed with "!", then the
will be considered invalid if the status code
This is mostly used to check for multiple
string <string> : test the exact string match in the HTTP response
A health check respose will be considered valid if
response's body contains this exact string. If
"string" keyword is prefixed with "!", then the
will be considered invalid if the body contains
string. This can be used to look for a mandatory word
the end of a dynamic page, or to detect a failure when
specific error appears on the check page (eg: a
trace).
trace).
rstring <regex> : test a regular expression on the HTTP response
A health check respose will be considered valid if
response's body matches this expression. If the
keyword is prefixed with "!", then the response will
considered invalid if the body matches the
This can be used to look for a mandatory word at the
of a dynamic page, or to detect a failure when a
error appears on the check page (eg: a stack
It is important to note that the responses will be limited to a certain
defined by the global "tune.chksize" option, which defaults to 16384
Thus, too large responses may not contain the mandatory pattern when
"string" or "rstring". If a large response is absolutely required, it
possible to change the default max size by setting the global
However, it is worth keeping in mind that parsing very large responses
waste some CPU cycles, especially when regular expressions are used, and
it is always better to focus the checks on smaller
Last, if "http-check expect" is combined with "http-check
then this last one has precedence when the server responds with
Examples
# only accept status 200 as
http-check expect status
# consider SQL errors as
http-check expect ! string SQL\
# consider status 5xx only as
http-check expect ! rstatus
# check that we have a correct hexadecimal tag before
http-check expect rstring
See also : "option httpchk", "http-check
http-check
Enable emission of a state header with HTTP health
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
When this option is set, haproxy will systematically send a special
"X-Haproxy-Server-State" with a list of parameters indicating to each
how they are seen by haproxy. This can be used for instance when a server
manipulated without access to haproxy and the operator needs to know
haproxy still sees it up or not, or if the server is the last one in a
The header is composed of fields delimited by semi-colons, the first of
is a word ("UP", "DOWN", "NOLB"), possibly followed by a number of
checks on the total number before transition, just as appears in the
interface. Next headers are in the form "<variable>=<value>", indicating
no specific order some values available in the stats interface
- a variable "name", containing the name of the backend followed by a
("/") then the name of the server. This can be used when a server
checked in multiple
- a variable "node" containing the name of the haproxy node, as set in
global "node" variable, otherwise the system's hostname if
- a variable "weight" indicating the weight of the server, a slash
and the total weight of the farm (just counting usable servers).
helps to know if other servers are available to handle the load when
one
- a variable "scur" indicating the current number of concurrent
on the server, followed by a slash ("/") then the total number
connections on all servers of the same
- a variable "qcur" indicating the current number of requests in
server's
Example of a header received by the application server
>>> X-Haproxy-Server-State: UP 2/3; name=bck/srv2; node=lb1; weight=1/2;
scur=13/22;
See also : "option httpchk", "http-check
http-request { allow | deny | auth [realm <realm>]
[ { if | unless } <condition>
Access control for Layer 7
May be used in sections: defaults | frontend | listen |
no | yes | yes |
These set of options allow to fine control access to
frontend/listen/backend. Each option may be followed by if/unless and
First option with matched condition (or option without condition) is
For "deny" a 403 error will be returned, for "allow" normal processing
performed, for "auth" a 401/407 error code is returned so the
should be asked to enter a username and
There is no fixed limit to the number of http-request statements
instance.
Example:
instance.
Example:
acl nagios src 192.168.129.
acl local_net src 192.168.0.
acl auth_ok
http-request allow if
http-request allow if local_net
http-request auth realm Gimme if local_net
http-request
Example:
Example:
acl auth_ok http_auth_group(L1)
http-request auth unless
See also : "stats http-request", section 3.4 about userlists and section
about ACL
http-send-name-header
Add the server name to a request. Use the header string given by
May be used in sections: defaults | frontend | listen |
yes | no | yes |
Arguments
<header> The header string to use to send the server
The "http-send-name-header" statement causes the name of the
server to be added to the headers of an HTTP request. The
is added with the header string
See also :
id
Set a persistent ID to a
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments :
Set a persistent ID for the proxy. This ID must be unique and
An unused ID will automatically be assigned if unset. The first
value will be 1. This ID is currently only returned in
ignore-persist { if | unless }
Declare a condition to ignore
May be used in sections: defaults | frontend | listen |
no | yes | yes |
By default, when cookie persistence is enabled, every requests
the cookie are unconditionally persistent (assuming the target server is
and
The "ignore-persist" statement allows one to declare various
conditions which, when met, will cause a request to ignore
This is sometimes useful to load balance requests for static files,
oftenly don't require persistence. This can also be used to fully
persistence for a specific User-Agent (for example, some web crawler
Combined with "appsession", it can also help reduce HAProxy memory usage,
the appsession table won't grow if persistence is
The persistence is ignored when an "if" condition is met, or unless
"unless" condition is
See also : "force-persist", "cookie", and section 7 about ACL
log
log <address> <facility> [<level>
Enable per-instance logging of events and
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
global should be used when the instance's logging parameters are
same as the global ones. This is the most common usage.
replaces <address>, <facility> and <level> with those of the
entries found in the "global" section. Only one "log
statement may be used per instance, and this form takes no
parameter.
parameter.
<address> indicates where to send the logs. It takes the same format
for the "global" section's logs, and can be one of
- An IPv4 address optionally followed by a colon (':') and a
port. If no port is specified, 514 is used by default
standard syslog
- A filesystem path to a UNIX domain socket, keeping in
considerations for chroot (be sure the path is
inside the chroot) and uid/gid (be sure the path
appropriately
<facility> must be one of the 24 standard syslog facilities
kern user mail daemon auth syslog lpr
uucp cron auth2 ftp ntp audit alert
local0 local1 local2 local3 local4 local5 local6
<level> is optional and can be specified to filter outgoing messages.
default, all messages are sent. If a level is specified,
messages with a severity at least as important as this
will be sent. An optional minimum level can be specified. If
is set, logs emitted with a more severe level than this one
be capped to this level. This is used to avoid sending
messages on all terminals on some default syslog
Eight levels are known
emerg alert crit err warning notice info
Note that up to two "log" entries may be specified per instance. However,
"log global" is used and if the "global" section already contains 2
entries, then additional log entries will be
Also, it is important to keep in mind that it is the frontend which
what to log from a connection, and that in case of content switching, the
entries from the backend will be ignored. Connections are logged at
"info".
"info".
However, backend log declaration define how and where servers status
will be logged. Level "notice" will be used to indicate a server going
"warning" will be used for termination signals and definitive
termination, and "alert" will be used for when a server goes
Note : According to RFC3164, messages are truncated to 1024 bytes
being
Example
log
log 127.0.0.1:514 local0 notice # only send important
log 127.0.0.1:514 local0 notice notice # same but limit output
maxconn
Fix the maximum number of concurrent connections on a
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<conns> is the maximum number of concurrent connections the frontend
accept to serve. Excess connections will be queued by the
in the socket's listen queue and will be served once a
closes.
closes.
If the system supports it, it can be useful on big sites to raise this
very high so that haproxy manages connection queues, instead of leaving
clients with unanswered connection attempts. This value should not exceed
global maxconn. Also, keep in mind that a connection contains two
of 8kB each, as well as some other data resulting in about 17 kB of RAM
consumed per established connection. That means that a medium system
with 1GB of RAM can withstand around 40000-50000 concurrent connections
properly
Also, when <conns> is set to large values, it is possible that the
are not sized to accept such loads, and for this reason it is generally
to assign them some reasonable connection
By default, this value is set to
See also : "server", global section's "maxconn",
mode { tcp|http|health
Set the running mode or protocol of the
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
tcp The instance will work in pure TCP mode. A full-duplex
will be established between clients and servers, and no layer
examination will be performed. This is the default mode.
should be used for SSL, SSH,
http The instance will work in HTTP mode. The client request will
analyzed in depth before connecting to any server. Any
which is not RFC-compliant will be rejected. Layer 7
processing and switching will be possible. This is the mode
brings HAProxy most of its
health The instance will work in "health" mode. It will just reply
to incoming connections and close the connection. Nothing will
logged. This mode is used to reply to external components
checks. This mode is deprecated and should not be used anymore
it is possible to do the same and even better by combining TCP
HTTP modes with the "monitor"
When doing content switching, it is mandatory that the frontend and
backend are in the same mode (generally HTTP), otherwise the
will be
Example
defaults
mode
See also : "monitor",
monitor fail { if | unless }
Add a condition to report a failure to a monitor HTTP
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
if <cond> the monitor request will fail if the condition is
and will succeed otherwise. The condition should describe
combined test which must induce a failure if all
are met, for instance a low number of servers both in
backend and its
unless <cond> the monitor request will succeed only if the condition
satisfied, and will fail otherwise. Such a condition may
based on a test on the presence of a minimum number of
servers in a list of
This statement adds a condition which can force the response to a
request to report a failure. By default, when an external component
the URI dedicated to monitoring, a 200 response is returned. When one of
conditions above is met, haproxy will return 503 instead of 200. This
very useful to report a site failure to an external component which may
routing advertisements between multiple sites on the availability reported
haproxy. In this case, one would rely on an ACL involving the
criterion. Note that "monitor fail" only works in HTTP mode. Both
messages may be tweaked using "errorfile" or "errorloc" if
Example:
Example:
frontend
mode
acl site_dead nbsrv(dynamic) lt
acl site_dead nbsrv(static) lt
monitor-uri
monitor fail if
See also : "monitor-net", "monitor-uri", "errorfile",
monitor-net
Declare a source network which is limited to monitor
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<source> is the source IPv4 address or network which will only be able
get monitor responses to any request. It can be either an
address, a host name, or an address followed by a slash
followed by a
In TCP mode, any connection coming from a source matching <source> will
the connection to be immediately closed without any log. This allows
equipment to probe the port and verify that it is still listening,
forwarding the connection to a remote
In HTTP mode, a connection coming from a source matching <source> will
accepted, the following response will be sent without waiting for a
then the connection will be closed : "HTTP/1.0 200 OK". This is
enough for any front-end HTTP probe to detect that the service is UP
running without forwarding the request to a backend
Monitor requests are processed very early. It is not possible to block
divert them using ACLs. They cannot be logged either, and it is the
purpose. They are only used to report HAProxy's health to an upper
nothing more. Right now, it is not possible to set failure conditions
requests caught by
Last, please note that only one "monitor-net" statement can be specified
a frontend. If more than one is found, only the last one will be
Example
# addresses .252 and .253 are just probing
frontend
monitor-net 192.168.0.
See also : "monitor fail",
monitor-uri
Intercept a URI used by external components' monitor
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<uri> is the exact URI which we want to intercept to return
health status instead of forwarding the
When an HTTP request referencing <uri> will be received on a
HAProxy will not forward it nor log it, but instead will return
"HTTP/1.0 200 OK" or "HTTP/1.0 503 Service unavailable", depending on
conditions defined with "monitor fail". This is normally enough for
front-end HTTP probe to detect that the service is UP and running
forwarding the request to a backend server. Note that the HTTP method,
version and all headers are ignored, but the request must at least be
at the HTTP level. This keyword may only be used with an HTTP-mode
Monitor requests are processed very early. It is not possible to block
divert them using ACLs. They cannot be logged either, and it is the
purpose. They are only used to report HAProxy's health to an upper
nothing more. However, it is possible to add any number of conditions
"monitor fail" and ACLs so that the result can be adjusted to whatever
can be imagined (most often the number of available servers in a
Example
# Use /haproxy_test to report haproxy's
frontend
mode
monitor-uri
See also : "monitor fail",
option
no option
Enable or disable early dropping of aborted requests pending in
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
In presence of very high loads, the servers will take some time to
The per-instance connection queue will inflate, and the response time
increase respective to the size of the queue times the average
response time. When clients will wait for more than a few seconds, they
often hit the "STOP" button on their browser, leaving a useless request
the queue, and slowing down other users, and the servers as well, because
request will eventually be served, then aborted at the first
encountered while delivering the
As there is no way to distinguish between a full STOP and a simple
close on the client side, HTTP agents should be conservative and
that the client might only have closed its output channel while waiting
the response. However, this introduces risks of congestion when lots of
do the same, and is completely useless nowadays because probably no client
all will close the session while waiting for the response. Some HTTP
support this behaviour (Squid, Apache, HAProxy), and others do not (TUX,
hardware-based load balancers). So the probability for a closed input
to represent a user hitting the "STOP" button is close to 100%, and the
of being the single component to break rare but valid traffic is
low, which adds to the temptation to be able to abort a session early
still not served and not pollute the
In HAProxy, the user can choose the desired behaviour using the
"abortonclose". By default (without the option) the behaviour is
compliant and aborted requests will be served. But when the option
specified, a session with an incoming channel closed will be aborted
it is still possible, either pending in the queue for a connection slot,
during the connection establishment if the server has not yet
the connection request. This considerably reduces the queue size and the
on saturated servers when users are tempted to click on STOP, which in
reduces the response time for other
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "timeout queue" and server's "maxconn" and "maxqueue"
option
no option
Enable or disable relaxing of HTTP request
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
By default, HAProxy complies with RFC2616 in terms of message parsing.
means that invalid characters in header names are not permitted and cause
error to be returned to the client. This is the desired behaviour as
forbidden characters are essentially used to build attacks exploiting
weaknesses, and bypass security filtering. Sometimes, a buggy browser
server will emit invalid header names for whatever reason
implementation) and the issue will not be immediately fixed. In such a
it is possible to relax HAProxy's header name parser to accept any
even if that does not make sense, by specifying this
This option should never be enabled by default as it hides application
and open security breaches. It should only be deployed after a problem
been
When this option is enabled, erroneous header names will still be accepted
requests, but the complete request will be captured in order to permit
analysis using the "show errors" request on the UNIX stats socket. Doing
also helps confirming that the issue has been
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option accept-invalid-http-response" and "show errors" on
stats
option
no option
Enable or disable relaxing of HTTP response
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
By default, HAProxy complies with RFC2616 in terms of message parsing.
means that invalid characters in header names are not permitted and cause
error to be returned to the client. This is the desired behaviour as
forbidden characters are essentially used to build attacks exploiting
weaknesses, and bypass security filtering. Sometimes, a buggy browser
server will emit invalid header names for whatever reason
implementation) and the issue will not be immediately fixed. In such a
it is possible to relax HAProxy's header name parser to accept any
even if that does not make sense, by specifying this
This option should never be enabled by default as it hides application
and open security breaches. It should only be deployed after a problem
been
When this option is enabled, erroneous header names will still be accepted
responses, but the complete response will be captured in order to
later analysis using the "show errors" request on the UNIX stats
Doing this also helps confirming that the issue has been
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option accept-invalid-http-request" and "show errors" on
stats
option
no option
Use either all backup servers at a time or only the first
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
By default, the first operational backup server gets all traffic when
servers are all down. Sometimes, it may be preferred to use multiple
at once, because one will not be enough. When "option allbackups" is
the load balancing will be performed among all backup servers when all
ones are unavailable. The same load balancing algorithm will be used and
servers' weights will be respected. Thus, there will not be any
order between the backup servers
This option is mostly used with static server farms dedicated to return
"sorry" page when an application is completely
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
option
no option
Analyze all server responses and block requests with cacheable
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
Some high-level frameworks set application cookies everywhere and do
always let enough control to the developer to manage how the responses
be cached. When a session cookie is returned on a cacheable object, there is
high risk of session crossing or stealing between users traversing the
caches. In some situations, it is better to block the response than to
some sensitive session information go in the
The option "checkcache" enables deep inspection of all server responses
strict compliance with HTTP specification in terms of cacheability.
carefully checks "Cache-control", "Pragma" and "Set-cookie" headers in
response to check if there's a risk of caching a cookie on a
proxy. When this option is enabled, the only responses which can be
to the client are
- all those without "Set-Cookie" header
- all those with a return code other than 200, 203, 206, 300, 301,
provided that the server has not set a "Cache-control: public" header
- all those that come from a POST request, provided that the server has
set a 'Cache-Control: public' header
- those with a 'Pragma: no-cache'
- those with a 'Cache-control: private'
- those with a 'Cache-control: no-store'
- those with a 'Cache-control: max-age=0'
- those with a 'Cache-control: s-maxage=0'
- those with a 'Cache-control: no-cache'
- those with a 'Cache-control: no-cache="set-cookie"'
- those with a 'Cache-control: no-cache="set-cookie,'
(allowing other fields after
If a response doesn't respect these requirements, then it will be
just as if it was from an "rspdeny" filter, with an "HTTP 502 bad
The session state shows "PH--" meaning that the proxy blocked the
during headers processing. Additionally, an alert will be sent in the logs
that admins are informed that there's something to be
Due to the high impact on the application, the application should be
in depth with the option enabled before going to production. It is also
good practice to always activate it during tests, even if it is not used
production, as it will report potentially dangerous application
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
option
no option
Enable or disable the sending of TCP keepalive packets on the client
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
When there is a firewall or any session-aware component between a client
a server, and when the protocol involves very long sessions with long
periods (eg: remote desktops), there is a risk that one of the
components decides to expire a session which has remained idle for too
Enabling socket-level TCP keep-alives makes the system regularly send
to the other end of the connection, leaving it active. The delay
keep-alive probes is controlled by the system only and depends both on
operating system and its tuning
It is important to understand that keep-alive packets are neither emitted
received at the application level. It is only the network stacks which
them. For this reason, even if one side of the proxy already uses
to maintain its connection alive, those keep-alive packets will not
forwarded to the other side of the
Please note that this has nothing to do with HTTP
Using option "clitcpka" enables the emission of TCP keep-alive probes on
client side of a connection, which should help when session expirations
noticed between HAProxy and a
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option srvtcpka", "option
option
Enable continuous traffic statistics
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
By default, counters used for statistics calculation are
only when a session finishes. It works quite well when serving
objects, but with big ones (for example large images or archives)
with A/V streaming, a graph generated from haproxy counters looks
a hedgehog. With this option enabled counters get incremented
during a whole session. Recounting touches a hotpath directly
it is not enabled by default, as it has small performance impact (~0.
option
no option
Enable or disable logging of normal, successful
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
There are large sites dealing with several thousand connections per
and for which logging is a major pain. Some of them are even forced to
logs off and cannot debug production issues. Setting this option ensures
normal connections, those which experience no error, no timeout, no retry
redispatch, will not be logged. This leaves disk space for anomalies. In
mode, the response status code is checked and return codes 5xx will still
logged.
logged.
It is strongly discouraged to use this option as most of the time, the key
complex issues is in the normal logs which will not be logged here. If
need to separate logs, see the "log-separate-errors" option
See also : "log", "dontlognull", "log-separate-errors" and section 8
logging.
logging.
option
no option
Enable or disable logging of null
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
In certain environments, there are components which will regularly connect
various systems to ensure that they are still alive. It can be the case
another load balancer as well as from monitoring systems. By default, even
simple port probe or scan will produce a log. If those connections
the logs too much, it is possible to enable option "dontlognull" to
that a connection on which no data has been transferred will not be
which typically corresponds to those
It is generally recommended not to use this option in
environments (eg: internet), otherwise scans and other malicious
would not be
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "log", "monitor-net", "monitor-uri" and section 8 about
option
no option
Enable or disable active connection closing after response is
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
Some HTTP servers do not necessarily close the connections when they
the "Connection: close" set by "option httpclose", and if the client does
close either, then the connection remains open till the timeout expires.
causes high number of simultaneous connections on the servers and shows
global session times in the
When this happens, it is possible to use "option forceclose". It
actively close the outgoing server channel as soon as the server has
to respond. This option implicitly enables the "httpclose" option. Note
this option also enables the parsing of the full request and response,
means we can close the connection to the server very quickly, releasing
resources earlier than with
This option may also be combined with "option http-pretend-keepalive",
will disable sending of the "Connection: close" header, but will still
the connection to be closed once the whole response is
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option httpclose" and "option
option forwardfor [ except <network> ] [ header <name> ] [ if-none
Enable insertion of the X-Forwarded-For header to requests sent to
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<network> is an optional argument used to disable this option for
matching
<name> an optional argument to specify a different
header
Since HAProxy works in reverse-proxy mode, the servers see its IP address
their client address. This is sometimes annoying when the client's IP
is expected in server logs. To solve this problem, the well-known HTTP
"X-Forwarded-For" may be added by HAProxy to all requests sent to the
This header contains a value representing the client's IP address. Since
header is always appended at the end of the existing header list, the
must be configured to always use the last occurrence of this header only.
the server's manual to find how to enable use of this standard header.
that only the last occurrence of the header must be used, since it is
possible that the client has already brought
The keyword "header" may be used to supply a different header name to
the default "X-Forwarded-For". This can be useful where you might
have a "X-Forwarded-For" header from a different application (eg:
and you need preserve it. Also if your backend server doesn't use
"X-Forwarded-For" header and requires different one (eg: Zeus Web
require
Sometimes, a same HAProxy instance may be shared between a direct
access and a reverse-proxy access (for instance when an SSL reverse-proxy
used to decrypt HTTPS traffic). It is possible to disable the addition of
header for a known source address or network by adding the "except"
followed by the network address. In this case, any source IP matching
network will not cause an addition of this header. Most common uses are
private networks or 127.0.0.
Alternatively, the keyword "if-none" states that the header will only
added if it is not present. This should only be used in perfectly
environment, as this might cause a security issue if headers reaching
are under the control of the
This option may be specified either in the frontend or in the backend. If
least one of them uses it, the header will be added. Note that the
setting of the header subargument takes precedence over the frontend's
both are defined. In the case of the "if-none" argument, if at least one
the frontend or the backend does not specify it, it wants the addition to
mandatory, so it
It is important to note that by default, HAProxy works in tunnel mode
only inspects the first request of a connection, meaning that only the
request will have the header appended, which is certainly not what you
In order to fix this, ensure that any of the "httpclose", "forceclose"
"http-server-close" options is set when using this
Examples
# Public HTTP address also used by stunnel on the same
frontend
mode
option forwardfor except 127.0.0.1 # stunnel already adds the
# Those servers want the IP Address in
backend
mode
option forwardfor header
See also : "option httpclose", "option
"option
option
no option
Instruct the system to favor low interactive delays over performance in
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
In HTTP, each payload is unidirectional and has no notion of
Any agent is expected to queue data somewhat for a reasonably low
There are some very rare server-to-server applications that abuse the
protocol and expect the payload phase to be highly interactive, with
interleaved data chunks in both directions within a single request. This
absolutely not supported by the HTTP specification and will not work
most proxies or servers. When such applications attempt to do this
haproxy, it works but they will experience high delays due to the
optimizations which favor performance by instructing the system to wait
enough data to be available in order to only send full packets.
delays are around 200 ms per round trip. Note that this only happens
abnormal uses. Normal uses such as CONNECT requests nor WebSockets are
affected.
affected.
When "option http-no-delay" is present in either the frontend or the
used by a connection, all such optimizations will be disabled in order
make the exchanges as fast as possible. Of course this offers no guarantee
the functionality, as it may break at any other place. But if it works
HAProxy, it will work as fast as possible. This option should never be
by default, and should never be used at all unless such a buggy
is discovered. The impact of using this option is an increase of
usage and CPU usage, which may significantly lower performance in
latency
option
no option
Define whether haproxy will announce keepalive to the server or
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
When running with "option http-server-close" or "option forceclose",
adds a "Connection: close" header to the request forwarded to the
Unfortunately, when some servers see this header, they automatically
from using the chunked encoding for responses of unknown length, while
is totally unrelated. The immediate effect is that this prevents haproxy
maintaining the client connection alive. A second effect is that a client
a cache could receive an incomplete response without being aware of it,
consider the response
By setting "option http-pretend-keepalive", haproxy will make the
believe it will keep the connection alive. The server will then not fall
to the abnormal undesired above. When haproxy gets the whole response,
will close the connection with the server just as it would do with
"forceclose" option. That way the client gets a normal response and
connection is correctly closed on the server
It is recommended not to enable this option by default, because most
will more efficiently close the connection themselves after the last
and release its buffers slightly earlier. Also, the added packet on
network could slightly reduce the overall peak performance. However it
worth noting that when this option is enabled, haproxy will have
less work to do. So if haproxy is the bottleneck on the whole
enabling this option might save a few CPU
This option may be set both in a frontend and in a backend. It is enabled
at least one of the frontend or backend holding a connection has it
This option may be compbined with "option httpclose", which will
keepalive to be announced to the server and close to be announced to
client. This practice is discouraged
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option forceclose" and "option
option
no option
Enable or disable HTTP connection closing on the server
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
By default, when a client communicates with a server, HAProxy will
analyze, log, and process the first request of each connection.
"option http-server-close" enables HTTP connection-close mode on the
side while keeping the ability to support HTTP keep-alive and pipelining
the client side. This provides the lowest latency on the client side
network) and the fastest session reuse on the server side to save
resources, similarly to "option forceclose". It also permits
capable servers to be served in keep-alive mode to the clients if
conform to the requirements of RFC2616. Please note that some servers do
always conform to those requirements when they see "Connection: close" in
request. The effect will be that keep-alive will never be used. A
consists in enabling "option
At the moment, logs will not indicate whether requests came from the
session or not. The accept date reported in the logs corresponds to the
of the previous request, and the request time corresponds to the time
waiting for a new request. The keep-alive request time is still bound to
timeout defined by "timeout http-keep-alive" or "timeout http-request"
not
This option may be set both in a frontend and in a backend. It is enabled
at least one of the frontend or backend holding a connection has it
It is worth noting that "option forceclose" has precedence over
http-server-close" and that combining "http-server-close" with
basically achieve the same result as
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option forceclose", "option
"option httpclose" and "1.1. The HTTP transaction
option
no option
Make use of non-standard Proxy-Connection header instead of
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
While RFC2616 explicitly states that HTTP/1.1 agents must use
Connection header to indicate their wish of persistent or
connections, both browsers and proxies ignore this header for
connections and make use of the undocumented, non-standard
header instead. The issue begins when trying to put a load balancer
browsers and such proxies, because there will be a difference between
haproxy understands and what the client and the proxy agree
By setting this option in a frontend, haproxy can automatically switch to
that non-standard header if it sees proxied requests. A proxied request
defined here as one where the URI begins with neither a '/' nor a '*'.
choice of header only affects requests passing through proxies making use
one of the "httpclose", "forceclose" and "http-server-close" options.
that this option can only be specified in a frontend and will affect
request along its whole
Also, when this option is set, a request which requires authentication
automatically switch to use proxy authentication headers if it is itself
proxied request. That makes it possible to check or enforce authentication
front of an existing
This option should normally never be used, except in front of a
See also : "option httpclose", "option forceclose" and
http-server-close".
http-server-close".
option
option httpchk
option httpchk <method>
option httpchk <method> <uri>
Enable HTTP protocol to check on the servers
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<method> is the optional HTTP method used with the requests. When not
the "OPTIONS" method is used, as it generally requires low
processing and is easy to filter out from the logs. Any
may be used, though it is not recommended to invent
ones.
ones.
<uri> is the URI referenced in the HTTP requests. It defaults to " /
which is accessible by default on almost any server, but may
changed to any other URI. Query strings are
<version> is the optional HTTP version string. It defaults to "HTTP/1.
but some servers might behave incorrectly in HTTP 1.0, so
it to HTTP/1.1 may sometimes help. Note that the Host field
mandatory in HTTP/1.1, and as a trick, it is possible to pass
after "\r\n" following the version
By default, server health checks only consist in trying to establish a
connection. When "option httpchk" is specified, a complete HTTP request
sent once the TCP connection is established, and responses 2xx and 3xx
considered valid, while all other ones indicate a server failure,
the lack of any
The port and interval are specified in the server
This option does not necessarily require an HTTP backend, it also works
plain TCP backends. This is particularly useful to check simple scripts
to some dedicated ports using the inetd
Examples
# Relay HTTPS traffic to Apache instance and check service
# using HTTP request "OPTIONS * HTTP/1.1" on port
backend
mode
option httpchk OPTIONS * HTTP/1.1\r\nHost:\
server apache1 192.168.1.1:443 check port
See also : "option ssl-hello-chk", "option smtpchk", "option
"http-check" and the "check", "port" and "inter" server
option
no option
Enable or disable passive HTTP connection
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
By default, when a client communicates with a server, HAProxy will
analyze, log, and process the first request of each connection. If
httpclose" is set, it will check if a "Connection: close" header is
set in each direction, and will add one if missing. Each end should react
this by actively closing the TCP connection after each transfer,
resulting in a switch to the HTTP close mode. Any "Connection"
different from "close" will also be
It seldom happens that some servers incorrectly ignore this header and do
close the connection eventhough they reply "Connection: close". For
reason, they are not compatible with older HTTP 1.0 browsers. If this
it is possible to use the "option forceclose" which actively closes
request connection once the server responds. Option "forceclose"
releases the server connection earlier because it does not have to wait
the client to acknowledge
This option may be set both in a frontend and in a backend. It is enabled
at least one of the frontend or backend holding a connection has it
If "option forceclose" is specified too, it has precedence over
If "option http-server-close" is enabled at the same time as "httpclose",
basically achieves the same result as "option
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option forceclose", "option http-server-close"
"1.1. The HTTP transaction
option httplog [ clf
Enable logging of HTTP request, session state and
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
clf if the "clf" argument is added, then the output format will
the CLF format instead of HAProxy's default HTTP format. You
use this when you need to feed HAProxy's logs through a
log analyser which only support the CLF format and which is
extensible.
extensible.
By default, the log output format is very poor, as it only contains
source and destination addresses, and the instance name. By
"option httplog", each log line turns into a much richer format
but not limited to, the HTTP request, the connection timers, the
status, the connections numbers, the captured headers and cookies,
frontend, backend and server name, and of course the source address
ports.
ports.
This option may be set either in the frontend or the
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before it.
only "option httplog" will automatically clear the 'clf' mode if it was
by
See also : section 8 about
option
no option
Enable or disable plain HTTP proxy
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
It sometimes happens that people need a pure HTTP proxy which
basic proxy requests without caching nor any fancy feature. In this
it may be worth setting up an HAProxy instance with the "option
set. In this mode, no server is declared, and the connection is forwarded
the IP address and port found in the URL after the "http://"
No host address resolution is performed, so this only works when pure
addresses are passed. Since this option's usage perimeter is rather
it will probably be used only by experts who know they need exactly it.
if the clients are susceptible of sending keep-alive requests, it will
needed to add "option httpclose" to ensure that all requests will
be
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
Example
# this backend understands HTTP proxy requests and forwards them
backend
option
option
See also : "option
option
no option
Enable or disable independant timeout processing for both
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
By default, when data is sent over a socket, both the write timeout and
read timeout for that socket are refreshed, because we consider that there
activity on that socket, and we have no other means of guessing if we
receive data or
While this default behaviour is desirable for almost all applications,
exists a situation where it is desirable to disable it, and only refresh
read timeout if there are incoming data. This happens on sessions with
timeouts and low amounts of exchanged data such as telnet session. If
server suddenly disappears, the output data accumulates in the
socket buffers, both timeouts are correctly refreshed, and there is no
to know the server does not receive them, so we don't timeout. However,
the underlying protocol always echoes sent data, it would be enough by
to detect the issue using the read timeout. Note that this problem does
happen with more verbose protocols because data won't accumulate long in
socket
When this option is set on the frontend, it will disable read timeout
on data sent to the client. There probably is little use of this case.
the option is set on the backend, it will disable read timeout updates
data sent to the server. Doing so will typically break large HTTP posts
slow lines, so use it with
See also : "timeout client" and "timeout
option
Use LDAPv3 health checks for server
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
It is possible to test that the server correctly talks LDAPv3 instead of
testing that it accepts the TCP connection. When this option is set,
LDAPv3 anonymous simple bind message is sent to the server, and the
is analyzed to find an LDAPv3 bind response
The server is considered valid only when the LDAP response contains
resultCode (http://tools.ietf.org/html/rfc4511#section-4.1.
Logging of bind requests is server dependent see your documentation how
configure
Example
option
See also : "option
option
no option
Enable or disable logging of health
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
Enable health checks logging so it possible to check for example
was happening before a server crash. Failed health check are logged
server is UP and succeeded health checks if server is DOWN, so the
of additional information is
If health check logging is enabled no health check status is
when servers is set up
See also: "log" and section 8 about
option
no option
Change log level for non-completely successful
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
Sometimes looking for errors in logs is not easy. This option makes
raise the level of logs containing potentially interesting information
as errors, timeouts, retries, redispatches, or HTTP status codes 5xx.
level changes from "info" to "err". This makes it possible to log
separately to a different file with most syslog daemons. Be careful not
remove them from the original file, otherwise you would lose ordering
provides very important
Using this option, large sites dealing with several thousand connections
second may log normal traffic to a rotating buffer and only archive
error
See also : "log", "dontlognull", "dontlog-normal" and section 8
logging.
logging.
option
no option
Enable or disable early logging of HTTP
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
By default, HTTP requests are logged upon termination so that the
transfer time and the number of bytes appear in the logs. When large
are being transferred, it may take a while before the request appears in
logs. Using "option logasap", the request gets logged as soon as the
sends the complete headers. The only missing information in the logs will
the total number of bytes which will indicate everything except the
of data transferred, and the total time which will not take the
time into account. In such a situation, it's a good practice to capture
"Content-Length" response header so that the logs at least indicate how
bytes are expected to be
Examples
listen http_proxy 0.0.0.
mode
option
option
log 192.168.2.200
>>> Feb 6 12:14:14 localhost
haproxy[14389]: 10.0.1.2:33317 [06/Feb/2009:12:14:14.655] http-in
static/srv1 9/10/7/14/+30 200 +243 - - ---- 3/1/1/1/0 1/0
"GET /image.iso HTTP/1.
See also : "option httplog", "capture response header", and section 8
logging.
logging.
option mysql-check [ user <username>
Use MySQL health checks for server
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<username> This is the username which will be used when connecting to
server.
server.
If you specify a username, the check consists of sending two MySQL
one Client Authentication packet, and one QUIT packet, to correctly
MySQL session. We then parse the MySQL Handshake Initialisation packet
Error packet. It is a basic but useful test which does not produce error
aborted connect on the server. However, it requires adding an
in the MySQL table, like this
USE
INSERT INTO user (Host,User) values
FLUSH
If you don't specify a username (it is deprecated and not recommended),
check only consists in parsing the Mysql Handshake Initialisation packet
Error packet, we don't send anything in this mode. It was reported that
can generate lockout if check is too frequent and/or if there is not
traffic. In fact, you need in this case to check MySQL
value as if a connection is established successfully within fewer than
"max_connect_errors" attempts after a previous connection was
the error count for the host is cleared to zero. If HAProxy's server
blocked, the "FLUSH HOSTS" statement is the only way to unblock
Remember that this does not check database presence nor database
To do this, you can use an external check with xinetd for
The check requires MySQL >=3.22, for older version, please use TCP
Most often, an incoming MySQL server needs to see the client's IP address
various purposes, including IP privilege matching and connection
When possible, it is often wise to masquerade the client's IP address
connecting to the server using the "usesrc" argument of the "source"
which requires the cttproxy feature to be compiled in, and the MySQL
to route the client via the machine hosting
See also: "option
option
no option
Enable or disable immediate session resource cleaning after
May be used in sections: defaults | frontend | listen |
yes | yes | yes |
Arguments :
When clients or servers abort connections in a dirty way (eg: they
physically disconnected), the session timeouts triggers and the session
closed. But it will remain in FIN_WAIT1 state for some time in the
using some resources and possibly limiting the ability to establish
connections.
connections.
When this happens, it is possible to activate "option nolinger" which
the system to immediately remove any socket's pending data on close.
the session is instantly purged from the system's tables. This usually
side effects such as increased number of TCP resets due to old
getting immediately rejected. Some firewalls may sometimes complain
this
For this reason, it is not recommended to use this option when not
needed. You know that you need it when you have thousands of
sessions on your system (TIME_WAIT ones do not
This option may be used both on frontends and backends, depending on the
where it is required. Use it on the frontend for clients, and on the
for
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
option originalto [ except <network> ] [ header <name>
Enable insertion of the X-Original-To header to requests sent to
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<network> is an optional argument used to disable this option for
matching
<name> an optional argument to specify a different
header
Since HAProxy can work in transparent mode, every request from a client
be redirected to the proxy and HAProxy itself can proxy every request to
complex SQUID environment and the destination host from SO_ORIGINAL_DST
be lost. This is annoying when you want access rules based on destination
addresses. To solve this problem, a new HTTP header "X-Original-To" may
added by HAProxy to all requests sent to the server. This header contains
value representing the original destination IP address. Since this must
configured to always use the last occurrence of this header only. Note
only the last occurrence of the header must be used, since it is
possible that the client has already brought
The keyword "header" may be used to supply a different header name to
the default "X-Original-To". This can be useful where you might
have a "X-Original-To" header from a different application, and you
preserve it. Also if your backend server doesn't use the
header and requires different
Sometimes, a same HAProxy instance may be shared between a direct
access and a reverse-proxy access (for instance when an SSL reverse-proxy
used to decrypt HTTPS traffic). It is possible to disable the addition of
header for a known source address or network by adding the "except"
followed by the network address. In this case, any source IP matching
network will not cause an addition of this header. Most common uses are
private networks or 127.0.0.
This option may be specified either in the frontend or in the backend. If
least one of them uses it, the header will be added. Note that the
setting of the header subargument takes precedence over the frontend's
both are
It is important to note that by default, HAProxy works in tunnel mode
only inspects the first request of a connection, meaning that only the
request will have the header appended, which is certainly not what you
In order to fix this, ensure that any of the "httpclose", "forceclose"
"http-server-close" options is set when using this
Examples
# Original Destination
frontend
mode
option originalto except 127.0.0.
# Those servers want the IP Address in
backend
mode
option originalto header
See also : "option httpclose", "option
"option
option
no option
Enable or disable forced persistence on down
May be used in sections: defaults | frontend | listen |
yes | no | yes |
Arguments :
When an HTTP request reaches a backend with a cookie which references a
server, by default it is redispatched to another server. It is possible
force the request to be sent to the dead server first using "option
if absolutely needed. A common use case is when servers are under
load and spend their time flapping. In this case, the users would still
directed to the server they opened the session on, in the hope they would
correctly served. It is recommended to use "option redispatch" in
with this option so that in the event it would not be possible to connect
the server at all (server definitely dead), the client would finally
redirected to another valid
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option redispatch", "retries",
option
no option
Enable or disable session redistribution in case of connection
May be used in sections: defaults | frontend | listen |
yes | no | yes |
Arguments :
In HTTP mode, if a server designated by a cookie is down, clients
definitely stick to it because they cannot flush the cookie, so they will
be able to access the service
Specifying "option redispatch" will allow the proxy to break
persistence and redistribute them to a working
It also allows to retry last connection to another server in case of
connection failures. Of course, it requires having "retries" set to a
value.
value.
This form is the preferred form, which replaces both the "redispatch"
"redisp"
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "redispatch", "retries",
option
option smtpchk <hello>
Use SMTP health checks for server
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<hello> is an optional argument. It is the "hello" command to use. It
be either "HELO" (for SMTP) or "EHLO" (for ESTMP). All
values will be turned into the default command
<domain> is the domain name to present to the server. It may only
specified (and is mandatory) if the hello command has
specified. By default, "localhost" is
When "option smtpchk" is set, the health checks will consist in
connections followed by an SMTP command. By default, this command
"HELO localhost". The server's return code is analyzed and only return
starting with a "2" will be considered as valid. All other
including a lack of response will constitute an error and will indicate
dead
This test is meant to be used with SMTP servers or relays. Depending on
request, it is possible that some servers do not log each connection
so you may want to experiment to improve the behaviour. Using telnet on
25 is often easier than adjusting the
Most often, an incoming SMTP server needs to see the client's IP address
various purposes, including spam filtering, anti-spoofing and logging.
possible, it is often wise to masquerade the client's IP address
connecting to the server using the "usesrc" argument of the "source"
which requires the cttproxy feature to be compiled
Example
option smtpchk HELO mydomain.
See also : "option httpchk",
option
no option
Enable or disable collecting & providing separate statistics for each
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
option
no option
Enable or disable automatic kernel acceleration on sockets in both
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
When this option is enabled either on a frontend or on a backend,
will automatically evaluate the opportunity to use kernel tcp splicing
forward data between the client and the server, in either direction.
uses heuristics to estimate if kernel splicing might improve performance
not. Both directions are handled independently. Note that the heuristics
are not much aggressive in order to limit excessive use of splicing.
option requires splicing to be enabled at compile time, and may be
disabled with the global option "nosplice". Since splice uses pipes, using
requires that there are enough spare
Important note: kernel-based TCP splicing is a Linux-specific feature
first appeared in kernel 2.6.25. It offers kernel-based acceleration
transfer data between sockets without copying these data to user-space,
providing noticeable performance gains and CPU cycles savings. Since
early implementations are buggy, corrupt data and/or are inefficient,
feature is not enabled by default, and it should be used with extreme
While it is not possible to detect the correctness of an
2.6.29 is the first version offering a properly working implementation.
case of doubt, splicing may be globally disabled using the global
keyword.
keyword.
Example
option
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option splice-request", "option splice-response", and
options "nosplice" and
option
no option
Enable or disable automatic kernel acceleration on sockets for
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
When this option is enabled either on a frontend or on a backend,
will user kernel tcp splicing whenever possible to forward data going
the client to the server. It might still use the recv/send scheme if
are no spare pipes left. This option requires splicing to be enabled
compile time, and may be globally disabled with the global option
Since splice uses pipes, using it requires that there are enough spare
Important note: see "option splice-auto" for usage
Example
option
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option splice-auto", "option splice-response", and global
"nosplice" and
option
no option
Enable or disable automatic kernel acceleration on sockets for
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
When this option is enabled either on a frontend or on a backend,
will user kernel tcp splicing whenever possible to forward data going
the server to the client. It might still use the recv/send scheme if
are no spare pipes left. This option requires splicing to be enabled
compile time, and may be globally disabled with the global option
Since splice uses pipes, using it requires that there are enough spare
Important note: see "option splice-auto" for usage
Example
option
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option splice-auto", "option splice-request", and global
"nosplice" and
option
no option
Enable or disable the sending of TCP keepalive packets on the server
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
When there is a firewall or any session-aware component between a client
a server, and when the protocol involves very long sessions with long
periods (eg: remote desktops), there is a risk that one of the
components decides to expire a session which has remained idle for too
Enabling socket-level TCP keep-alives makes the system regularly send
to the other end of the connection, leaving it active. The delay
keep-alive probes is controlled by the system only and depends both on
operating system and its tuning
It is important to understand that keep-alive packets are neither emitted
received at the application level. It is only the network stacks which
them. For this reason, even if one side of the proxy already uses
to maintain its connection alive, those keep-alive packets will not
forwarded to the other side of the
Please note that this has nothing to do with HTTP
Using option "srvtcpka" enables the emission of TCP keep-alive probes on
server side of a connection, which should help when session expirations
noticed between HAProxy and a
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option clitcpka", "option
option
Use SSLv3 client hello health checks for server
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
When some SSL-based protocols are relayed in TCP mode through HAProxy, it
possible to test that the server correctly talks SSL instead of just
that it accepts the TCP connection. When "option ssl-hello-chk" is set,
SSLv3 client hello messages are sent once the connection is established
the server, and the response is analyzed to find an SSL server hello
The server is considered valid only when the response contains this
hello
All servers tested till there correctly reply to SSLv3 client hello
and most servers tested do not even log the requests containing only
messages, which is
See also: "option
option
no option
Enable or disable the saving of one ACK packet during the accept
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
When an HTTP connection request comes in, the system acknowledges it
behalf of HAProxy, then the client immediately sends its request, and
system acknowledges it too while it is notifying HAProxy about the
connection. HAProxy then reads the request and responds. This means that
have one TCP ACK sent by the system for nothing, because the request
very well be acknowledged by HAProxy when it sends its
For this reason, in HTTP mode, HAProxy automatically asks the system to
sending this useless ACK on platforms which support it (currently at
Linux). It must not cause any problem, because the system will send it
after 40 ms if the response takes more time than expected to
During complex network debugging sessions, it may be desirable to
this optimization because delayed ACKs can make troubleshooting more
when trying to identify where packets are delayed. It is then possible
fall back to normal behaviour by specifying "no option
It is also possible to force it for non-HTTP proxies by simply
"option tcp-smart-accept". For instance, it can make sense with some
such as SMTP where the server speaks
It is recommended to avoid forcing this option in a defaults section. In
of doubt, consider setting it back to automatic values by prepending
"default" keyword before it, or disabling it using the "no"
See also : "option
option
no option
Enable or disable the saving of one ACK packet during the connect
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
On certain systems (at least Linux), HAProxy can ask the kernel not
immediately send an empty ACK upon a connection request, but to
send the buffer request instead. This saves one packet on the network
thus boosts performance. It can also be useful for some servers, because
immediately get the request along with the incoming
This feature is enabled when "option tcp-smart-connect" is set in a
It is not enabled by default because it makes network troubleshooting
complex.
complex.
It only makes sense to enable it with protocols where the client speaks
such as HTTP. In other situations, if there is no data to send in place
the ACK, a normal ACK is
If this option has been enabled in a "defaults" section, it can be
in a specific instance by prepending the "no" keyword before
See also : "option
option
Enable or disable the sending of TCP keepalive packets on both
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
When there is a firewall or any session-aware component between a client
a server, and when the protocol involves very long sessions with long
periods (eg: remote desktops), there is a risk that one of the
components decides to expire a session which has remained idle for too
Enabling socket-level TCP keep-alives makes the system regularly send
to the other end of the connection, leaving it active. The delay
keep-alive probes is controlled by the system only and depends both on
operating system and its tuning
It is important to understand that keep-alive packets are neither emitted
received at the application level. It is only the network stacks which
them. For this reason, even if one side of the proxy already uses
to maintain its connection alive, those keep-alive packets will not
forwarded to the other side of the
Please note that this has nothing to do with HTTP
Using option "tcpka" enables the emission of TCP keep-alive probes on
the client and server sides of a connection. Note that this is
only in "defaults" or "listen" sections. If this option is used in
frontend, only the client side will get keep-alives, and if this option
used in a backend, only the server side will get keep-alives. For
reason, it is strongly recommended to explicitly use "option clitcpka"
"option srvtcpka" when the configuration is split between frontends
backends.
backends.
See also : "option clitcpka", "option
option
Enable advanced logging of TCP connections with session state and
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments :
By default, the log output format is very poor, as it only contains
source and destination addresses, and the instance name. By
"option tcplog", each log line turns into a much richer format including,
not limited to, the connection timers, the session status, the
numbers, the frontend, backend and server name, and of course the
address and ports. This option is useful for pure TCP proxies in order
find which of the client or server disconnects or times out. For normal
proxies, it's better to use "option httplog" which is even more
This option may be set either in the frontend or the
See also : "option httplog", and section 8 about
option
no option
Enable client-side transparent
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
This option was introduced in order to provide layer 7 persistence to layer
load balancers. The idea is to use the OS's ability to redirect an
connection for a remote address to a local process (here HAProxy), and
this process know what address was initially requested. When this option
used, sessions without cookies will be forwarded to the original
IP address of the incoming request (which should match that of
equipment), while requests with cookies will still be forwarded to
appropriate
Note that contrary to a common belief, this option does NOT make
present the client's IP to the server when establishing the
See also: the "usesrc" argument of the "source" keyword, and
"transparent" option of the "bind"
persist
persist
Enable RDP cookie-based
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<name> is the optional name of the RDP cookie to check. If omitted,
default cookie name "msts" will be used. There currently is
valid reason to change this
This statement enables persistence based on an RDP cookie. The RDP
contains all information required to find the server in the list of
servers. So when this option is set in the backend, the request is
and if an RDP cookie is found, it is decoded. If it matches a known
which is still UP (or if "option persist" is set), then the connection
forwarded to this
Note that this only makes sense in a TCP backend, but for this to work,
frontend must have waited long enough to ensure that an RDP cookie is
in the request buffer. This is the same requirement as with the
load-balancing method. Thus it is highly recommended to put all statements
a single "listen"
Also, it is important to understand that the terminal server will emit
RDP cookie only if it is configured for "token redirection mode", which
that the "IP address redirection" option is
Example
listen
bind
# wait up to 5s for an RDP cookie in the
tcp-request inspect-delay
tcp-request content accept if
# apply RDP cookie
persist
# if server is unknown, let's balance on the same
# alternatively, "balance leastconn" may be useful
balance
server srv1 1.1.1.
server srv2 1.1.1.
See also : "balance rdp-cookie", "tcp-request" and the "req_rdp_cookie"
rate-limit sessions
Set a limit on the number of new sessions accepted per second on a
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<rate> The <rate> parameter is an integer designating the maximum
of new sessions per second to accept on the
When the frontend reaches the specified number of new sessions per second,
stops accepting new connections until the rate drops below the limit
During this time, the pending sessions will be kept in the socket's
(in system buffers) and haproxy will not even be aware that sessions
pending. When applying very low limit on a highly loaded service, it may
sense to increase the socket's backlog using the "backlog"
This feature is particularly efficient at blocking connection-based
or service abuse on fragile servers. Since the session rate is measured
millisecond, it is extremely accurate. Also, the limit applies
no delay is needed at all to detect the
Example : limit the connection rate on SMTP to 10 per second
listen
mode
bind
rate-limit sessions
server 127.0.0.
Note : when the maximum rate is reached, the frontend's status appears
"FULL" in the statistics, exactly as when it is
See also : the "backlog" keyword and the "fe_sess_rate" ACL
redirect location <to> [code <code>] <option> [{if | unless}
redirect prefix <to> [code <code>] <option> [{if | unless}
Return an HTTP redirection if/unless a condition is
May be used in sections : defaults | frontend | listen |
no | yes | yes |
If/unless the condition is matched, the HTTP request will lead to a
response. If no condition is specified, the redirect applies
Arguments
<to> With "redirect location", the exact value in <to> is placed
the HTTP "Location" header. In case of "redirect prefix",
"Location" header is built from the concatenation of <to> and
complete URI, including the query string, unless the
option is specified (see below). As a special case, if
equals exactly "/" in prefix mode, then nothing is
before the original URI. It allows one to redirect to the
URL.
URL.
<code> The code is optional. It indicates which type of HTTP
is desired. Only codes 301, 302 and 303 are supported, and 302
used if no code is specified. 301 means "Moved permanently",
a browser may cache the Location. 302 means "Moved
and means that the browser should not cache the redirection.
is equivalent to 302 except that the browser will fetch
location with a GET
<option> There are several options which can be specified to adjust
expected behaviour of a redirection
-
When this keyword is used in a prefix-based redirection, then
location will be set without any possible query-string, which is
for directing users to a non-secure page for instance. It has no
with a location-type
-
This keyword may be used in conjunction with "drop-query" to
users who use a URL not ending with a '/' to the same one with the
It can be useful to ensure that search engines will only see one
For this, a return code 301 is
- "set-cookie
A "Set-Cookie" header will be added with NAME (and optionally
to the response. This is sometimes used to indicate that a user
been seen, for instance to protect against some types of DoS. No
cookie option is added, so the cookie will be a session cookie.
that for a browser, a sole cookie name without an equal sign
different from a cookie with an equal
- "clear-cookie
A "Set-Cookie" header will be added with NAME (and optionally "="),
with the "Max-Age" attribute set to zero. This will tell the browser
delete this cookie. It is useful for instance on logout pages. It
important to note that clearing the cookie "NAME" will not remove
cookie set with "NAME=value". You have to clear the cookie "NAME="
that, because the browser makes the
Example: move the login URL only to
acl clear dst_port
acl secure dst_port
acl login_page url_beg
acl logout url_beg
acl uid_given url_reg
acl cookie_set hdr_sub(cookie)
redirect prefix https://mysite.com set-cookie SEEN=1 if
redirect prefix https://mysite.com if login_page
redirect prefix http://mysite.com drop-query if login_page
redirect location http://mysite.com/ if !login_page
redirect location / clear-cookie USERID= if
Example: send redirects for request for articles without a
acl missing_slash path_reg
redirect code 301 prefix / drop-query append-slash if
See section 7 about ACL
redisp
redispatch
Enable or disable session redistribution in case of connection
May be used in sections: defaults | frontend | listen |
yes | no | yes |
Arguments :
In HTTP mode, if a server designated by a cookie is down, clients
definitely stick to it because they cannot flush the cookie, so they will
be able to access the service
Specifying "redispatch" will allow the proxy to break their persistence
redistribute them to a working
It also allows to retry last connection to another server in case of
connection failures. Of course, it requires having "retries" set to a
value.
value.
This form is deprecated, do not use it in any new configuration, use the
"option redispatch"
See also : "option
reqadd <string> [{if | unless}
Add a header at the end of the HTTP
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<string> is the complete line to be added. Any space or known
must be escaped using a backslash ('\'). Please refer to
6 about HTTP header manipulation for more
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
A new line consisting in <string> followed by a line feed will be added
the last header of an HTTP
Header transformations only apply to traffic which passes through
and not to traffic generated by HAProxy, such as health-checks or
responses.
responses.
Example : add "X-Proto: SSL" to requests coming via port
acl is-ssl dst_port
reqadd X-Proto:\ SSL if
See also: "rspadd", section 6 about HTTP header manipulation, and section
about
reqallow <search> [{if | unless}
reqiallow <search> [{if | unless} <cond>] (ignore
Definitely allow an HTTP request if a line matches a regular
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<search> is the regular expression applied to HTTP headers and to
request line. This is an extended regular expression.
grouping is supported and no preliminary backslash is
Any space or known delimiter must be escaped using a
('\'). The pattern applies to a full line at a time.
"reqallow" keyword strictly matches case while
ignores
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
A request containing any line which matches extended regular
<search> will mark the request as allowed, even if any later test
result in a deny. The test applies both to the request line and to
headers. Keep in mind that URLs in request line are case-sensitive
header names are
It is easier, faster and more powerful to use ACLs to write access
Reqdeny, reqallow and reqpass should be avoided in new
Example
# allow www.* but refuse *.
reqiallow ^Host:\
reqideny ^Host:\ .*\.
See also: "reqdeny", "block", section 6 about HTTP header manipulation,
section 7 about
reqdel <search> [{if | unless}
reqidel <search> [{if | unless} <cond>] (ignore
Delete all headers matching a regular expression in an HTTP
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<search> is the regular expression applied to HTTP headers and to
request line. This is an extended regular expression.
grouping is supported and no preliminary backslash is
Any space or known delimiter must be escaped using a
('\'). The pattern applies to a full line at a time. The
keyword strictly matches case while "reqidel" ignores
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
Any header line matching extended regular expression <search> in the
will be completely deleted. Most common use of this is to remove
and/or dangerous headers or cookies from a request before passing it to
next
Header transformations only apply to traffic which passes through
and not to traffic generated by HAProxy, such as health-checks or
responses. Keep in mind that header names are not
Example
# remove X-Forwarded-For header and SERVER
reqidel ^X-Forwarded-For:.
reqidel ^Cookie:.
See also: "reqadd", "reqrep", "rspdel", section 6 about HTTP
manipulation, and section 7 about
reqdeny <search> [{if | unless}
reqideny <search> [{if | unless} <cond>] (ignore
Deny an HTTP request if a line matches a regular
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<search> is the regular expression applied to HTTP headers and to
request line. This is an extended regular expression.
grouping is supported and no preliminary backslash is
Any space or known delimiter must be escaped using a
('\'). The pattern applies to a full line at a time.
"reqdeny" keyword strictly matches case while "reqideny"
case.
case.
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
A request containing any line which matches extended regular
<search> will mark the request as denied, even if any later test
result in an allow. The test applies both to the request line and to
headers. Keep in mind that URLs in request line are case-sensitive
header names are
A denied request will generate an "HTTP 403 forbidden" response once
complete request has been parsed. This is consistent with what is
using
It is easier, faster and more powerful to use ACLs to write access
Reqdeny, reqallow and reqpass should be avoided in new
Example
# refuse *.local, then allow www.
reqideny ^Host:\ .*\.
reqiallow ^Host:\
See also: "reqallow", "rspdeny", "block", section 6 about HTTP
manipulation, and section 7 about
reqpass <search> [{if | unless}
reqipass <search> [{if | unless} <cond>] (ignore
Ignore any HTTP request line matching a regular expression in next
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<search> is the regular expression applied to HTTP headers and to
request line. This is an extended regular expression.
grouping is supported and no preliminary backslash is
Any space or known delimiter must be escaped using a
('\'). The pattern applies to a full line at a time.
"reqpass" keyword strictly matches case while "reqipass"
case.
case.
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
A request containing any line which matches extended regular
<search> will skip next rules, without assigning any deny or allow
The test applies both to the request line and to request headers. Keep
mind that URLs in request line are case-sensitive while header names are
It is easier, faster and more powerful to use ACLs to write access
Reqdeny, reqallow and reqpass should be avoided in new
Example
# refuse *.local, then allow www.*, but ignore "www.private.
reqipass ^Host:\ www.private\.
reqideny ^Host:\ .*\.
reqiallow ^Host:\
See also: "reqallow", "reqdeny", "block", section 6 about HTTP
manipulation, and section 7 about
reqrep <search> <string> [{if | unless}
reqirep <search> <string> [{if | unless} <cond>] (ignore
Replace a regular expression with a string in an HTTP request
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<search> is the regular expression applied to HTTP headers and to
request line. This is an extended regular expression.
grouping is supported and no preliminary backslash is
Any space or known delimiter must be escaped using a
('\'). The pattern applies to a full line at a time. The
keyword strictly matches case while "reqirep" ignores
<string> is the complete line to be added. Any space or known
must be escaped using a backslash ('\'). References to
pattern groups are possible using the common \N form, with
being a single digit between 0 and 9. Please refer to
6 about HTTP header manipulation for more
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
Any line matching extended regular expression <search> in the request
the request line and header lines) will be completely replaced with
Most common use of this is to rewrite URLs or domain names in "Host"
Header transformations only apply to traffic which passes through
and not to traffic generated by HAProxy, such as health-checks or
responses. Note that for increased readability, it is suggested to add
spaces between the request and the response. Keep in mind that URLs
request line are case-sensitive while header names are
Example
# replace "/static/" with "/" at the beginning of any request
reqrep ^([^\ :]*)\ /static/(.*) \1\
# replace "www.mydomain.com" with "www" in the host
reqirep ^Host:\ www.mydomain.com Host:\
See also: "reqadd", "reqdel", "rsprep", section 6 about HTTP
manipulation, and section 7 about
reqtarpit <search> [{if | unless}
reqitarpit <search> [{if | unless} <cond>] (ignore
Tarpit an HTTP request containing a line matching a regular
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<search> is the regular expression applied to HTTP headers and to
request line. This is an extended regular expression.
grouping is supported and no preliminary backslash is
Any space or known delimiter must be escaped using a
('\'). The pattern applies to a full line at a time.
"reqtarpit" keyword strictly matches case while
ignores
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
A request containing any line which matches extended regular
<search> will be tarpitted, which means that it will connect to nowhere,
be kept open for a pre-defined time, then will return an HTTP error 500
that the attacker does not suspect it has been tarpitted. The status 500
be reported in the logs, but the completion flags will indicate "PT".
delay is defined by "timeout tarpit", or "timeout connect" if the former
not
The goal of the tarpit is to slow down robots attacking servers
identifiable requests. Many robots limit their outgoing number of
and stay connected waiting for a reply which can take several minutes
come. Depending on the environment and attack, it may be
efficient at reducing the load on the network and
Examples
# ignore user-agents reporting any flavour of "Mozilla" or "MSIE",
# block all
reqipass ^User-Agent:\.
reqitarpit
# block bad
acl badguys src 10.1.0.3 172.16.13.
reqitarpit . if
See also: "reqallow", "reqdeny", "reqpass", section 6 about HTTP
manipulation, and section 7 about
retries
Set the number of retries to perform on a server after a connection
May be used in sections: defaults | frontend | listen |
yes | no | yes |
Arguments
<value> is the number of times a connection attempt should be retried
a server when a connection either is refused or times out.
default value is
It is important to understand that this value applies to the number
connection attempts, not full requests. When a connection has
been established to a server, there will be no more
In order to avoid immediate reconnections to a server which is
a turn-around timer of 1 second is applied before a retry
When "option redispatch" is set, the last retry may be performed on
server even if a cookie references a different
See also : "option
rspadd <string> [{if | unless}
Add a header at the end of the HTTP
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<string> is the complete line to be added. Any space or known
must be escaped using a backslash ('\'). Please refer to
6 about HTTP header manipulation for more
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
A new line consisting in <string> followed by a line feed will be added
the last header of an HTTP
Header transformations only apply to traffic which passes through
and not to traffic generated by HAProxy, such as health-checks or
responses.
responses.
See also: "reqadd", section 6 about HTTP header manipulation, and section
about
rspdel <search> [{if | unless}
rspidel <search> [{if | unless} <cond>] (ignore
Delete all headers matching a regular expression in an HTTP
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<search> is the regular expression applied to HTTP headers and to
response line. This is an extended regular expression,
parenthesis grouping is supported and no preliminary
is required. Any space or known delimiter must be escaped
a backslash ('\'). The pattern applies to a full line at a
The "rspdel" keyword strictly matches case while
ignores
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
Any header line matching extended regular expression <search> in the
will be completely deleted. Most common use of this is to remove
and/or sensitive headers or cookies from a response before passing it to
client.
client.
Header transformations only apply to traffic which passes through
and not to traffic generated by HAProxy, such as health-checks or
responses. Keep in mind that header names are not
Example
# remove the Server header from
reqidel ^Server:.
See also: "rspadd", "rsprep", "reqdel", section 6 about HTTP
manipulation, and section 7 about
rspdeny <search> [{if | unless}
rspideny <search> [{if | unless} <cond>] (ignore
Block an HTTP response if a line matches a regular
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<search> is the regular expression applied to HTTP headers and to
response line. This is an extended regular expression,
parenthesis grouping is supported and no preliminary
is required. Any space or known delimiter must be escaped
a backslash ('\'). The pattern applies to a full line at a
The "rspdeny" keyword strictly matches case while
ignores
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
A response containing any line which matches extended regular
<search> will mark the request as denied. The test applies both to
response line and to response headers. Keep in mind that header names are
case-sensitive.
case-sensitive.
Main use of this keyword is to prevent sensitive information leak and
block the response before it reaches the client. If a response is denied,
will be replaced with an HTTP 502 error so that the client never
any sensitive
It is easier, faster and more powerful to use ACLs to write access
Rspdeny should be avoided in new
Example
# Ensure that no content type matching ms-word will
rspideny ^Content-type:\.
See also: "reqdeny", "acl", "block", section 6 about HTTP header
and section 7 about
rsprep <search> <string> [{if | unless}
rspirep <search> <string> [{if | unless} <cond>] (ignore
Replace a regular expression with a string in an HTTP response
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<search> is the regular expression applied to HTTP headers and to
response line. This is an extended regular expression,
parenthesis grouping is supported and no preliminary
is required. Any space or known delimiter must be escaped
a backslash ('\'). The pattern applies to a full line at a
The "rsprep" keyword strictly matches case while
ignores
<string> is the complete line to be added. Any space or known
must be escaped using a backslash ('\'). References to
pattern groups are possible using the common \N form, with
being a single digit between 0 and 9. Please refer to
6 about HTTP header manipulation for more
<cond> is an optional matching condition built from ACLs. It makes
possible to ignore this rule when other conditions are not
Any line matching extended regular expression <search> in the response
the response line and header lines) will be completely replaced
<string>. Most common use of this is to rewrite Location
Header transformations only apply to traffic which passes through
and not to traffic generated by HAProxy, such as health-checks or
responses. Note that for increased readability, it is suggested to add
spaces between the request and the response. Keep in mind that header
are not
Example
# replace "Location: 127.0.0.1:8080" with "Location: www.mydomain.
rspirep ^Location:\ 127.0.0.1:8080 Location:\ www.mydomain.
See also: "rspadd", "rspdel", "reqrep", section 6 about HTTP
manipulation, and section 7 about
server <name> <address>[:port]
Declare a server in a
May be used in sections : defaults | frontend | listen |
no | no | yes |
Arguments
<name> is the internal name assigned to this server. This name
appear in logs and alerts. If "http-send-server-name"
set, it will be added to the request header sent to the
<address> is the IPv4 address of the server. Alternatively, a
hostname is supported, but this name will be resolved
start-up. If no address is specified in front of the port, or
address "0.0.0.0" is specified, then the address used will be
same as the one used by the incoming connection. This makes
possible to relay connections for many IPs on a different port
to perform transparent forwarding with connection
<ports> is an optional port specification. If set, all connections
be sent to this port. If unset, the same port the
connected to will be used. The port may also be prefixed by a
or a "-". In this case, the server's port will be determined
adding this value to the client's
<param*> is a list of parameters for this server. The "server"
accepts an important number of options and has a complete
dedicated to it. Please refer to section 5 for more
Examples
server first 10.1.1.1:1080 cookie first check inter
server second 10.1.1.2:1080 cookie second check inter
See also: "default-server", "http-send-name-header" and section 5
server
source <addr>[:<port>] [usesrc { <addr2>[:<port2>] | client | clientip }
source <addr>[:<port>] [usesrc { <addr2>[:<port2>] | hdr_ip(<hdr>[,<occ>]) }
source <addr>[:<port>] [interface
Set the source address for outgoing
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<addr> is the IPv4 address HAProxy will bind to before connecting to
server. This address is also used as a source for health
The default value of 0.0.0.0 means that the system will
the most appropriate address to reach its
<port> is an optional port. It is normally not needed but may be
in some very specific contexts. The default value of zero
the system will select a free port. Note that port ranges are
supported in the backend. If you want to force port ranges,
have to specify them on each "server"
<addr2> is the IP address to present to the server when connections
forwarded in full transparent proxy mode. This is currently
supported on some patched Linux kernels. When this address
specified, clients connecting to the server will be
with this address, while health checks will still use the
<addr>.
<addr>.
<port2> is the optional port to present to the server when
are forwarded in full transparent proxy mode (see <addr2>
The default value of zero means the system will select a
port.
port.
<hdr> is the name of a HTTP header in which to fetch the IP to bind
This is the name of a comma-separated header list which
contain multiple IP addresses. By default, the last occurrence
used. This is designed to work with the X-Forwarded-For
and to automatically bind to the the client's IP address as
by previous proxy, typically Stunnel. In order to use
occurrence from the last one, please see the <occ>
below. When the header (or occurrence) is not found, no
is performed so that the proxy's default IP address is used.
keep in mind that the header name is case insensitive, as for
HTTP
<occ> is the occurrence number of a value to be used in a
header. This is to be used in conjunction with
in order to specificy which occurrence to use for the source
address. Positive values indicate a position from the
occurrence, 1 being the first one. Negative values
positions relative to the last one, -1 being the last one.
is helpful for situations where an X-Forwarded-For header is
at the entry point of an infrastructure and must be used
proxy layers away. When this value is not specified, -1
assumed. Passing a zero here disables the
<name> is an optional interface name to which to bind to for
traffic. On systems supporting this features (currently,
Linux), this allows one to bind all traffic to the server
this interface even if it is not the one the system would
based on routing tables. This should be used with extreme
Note that using this option requires root
The "source" keyword is useful in complex environments where a
address only is allowed to connect to the servers. It may be needed when
private address must be used through a public gateway for instance, and it
known that the system cannot determine the adequate source address by
An extension which is available on certain patched Linux kernels may be
through the "usesrc" optional keyword. It makes it possible to connect to
servers with an IP address which does not belong to the system itself.
is called "full transparent proxy mode". For this to work, the
servers have to route their traffic back to this address through the
running HAProxy, and IP forwarding must generally be enabled on this
In this "full transparent proxy" mode, it is possible to force a specific
address to be presented to the servers. This is not much used in fact. A
common use is to tell HAProxy to present the client's IP address. For
there are two methods
- present the client's IP and port addresses. This is the most
mode, but it can cause problems when IP connection tracking is enabled
the machine, because a same connection may be seen twice with
states. However, this solution presents the huge advantage of
limiting the system to the 64k outgoing address+port couples, because
of the client ranges may be
- present only the client's IP address and select a spare port.
solution is still quite elegant but slightly less transparent
firewalls logs will not match upstream's). It also presents the
of limiting the number of concurrent connections to the usual 64k
However, since the upstream and downstream ports are different, local
connection tracking on the machine will not be upset by the reuse of
same
Note that depending on the transparent proxy technology used, it may
required to force the source address. In fact, cttproxy version 2 requires
IP address in <addr> above, and does not support setting of "0.0.0.0" as
IP address because it creates NAT entries which much match the exact
address. Tproxy version 4 and some other kernel patches which work in
forwarding mode generally will not have this
This option sets the default source for all servers in the backend. It
also be specified in a "defaults" section. Finer source address
is possible at the server level using the "source" server option. Refer
section 5 for more
Examples
backend
# Connect to the servers using our 192.168.1.200 source
source 192.168.1.
backend
# Connect to the SSL farm from the client's source
source 192.168.1.200 usesrc
backend
# Connect to the SSL farm from the client's source address and
# not recommended if IP conntrack is present on the local
source 192.168.1.200 usesrc
backend
# Connect to the SSL farm from the client's source address.
# is more
source 192.168.1.200 usesrc
backend
# Connect to the SMTP farm from the client's source
# with Tproxy version
source 0.0.0.0 usesrc
backend
# Connect to the servers using the client's IP as seen by
#
source 0.0.0.0 usesrc
See also : the "source" server option in section 5, the Tproxy patches
the Linux kernel on www.balabit.com, the "bind"
srvtimeout <timeout>
Set the maximum inactivity time on the server
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<timeout> is the timeout value specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
The inactivity timeout applies when the server is expected to acknowledge
send data. In HTTP mode, this timeout is particularly important to
during the first phase of the server's response, when it has to send
headers, as it directly represents the server's processing time for
request. To find out what value to put there, it's often good to start
what would be considered as unacceptable response times, then check the
to observe the response time distribution, and adjust the value
The value is specified in milliseconds by default, but can be in any
unit if the number is suffixed by the unit, as specified at the top of
document. In TCP mode (and to a lesser extent, in HTTP mode), it is
recommended that the client timeout remains equal to the server timeout
order to avoid complex situations to debug. Whatever the expected
response times, it is a good practice to cover at least one or several
packet losses by specifying timeouts that are slightly above multiples of
seconds (eg: 4 or 5 seconds
This parameter is specific to backends, but can be specified once for all
"defaults" sections. This is in fact one of the easiest solutions not
forget about it. An unspecified timeout results in an infinite timeout,
is not recommended. Such a usage is accepted and works but reports a
during startup because it may results in accumulation of expired sessions
the system if the system's timeouts are not configured
This parameter is provided for compatibility but is currently
Please use "timeout server"
See also : "timeout server", "timeout client" and
stats admin { if | unless }
Enable statistics admin level if/unless a condition is
May be used in sections : defaults | frontend | listen |
no | no | yes |
This statement enables the statistics admin level if/unless a condition
matched.
matched.
The admin level allows to enable/disable servers from the web interface.
default, statistics page is read-only for security
Note : Consider not using this feature in multi-process mode (nbproc >
unless you know what you do : memory is not shared between
processes, which can result in random
Currently, the POST request is limited to the buffer size minus the
buffer space, which means that if the list of servers is too long,
request won't be processed. It is recommended to alter few servers at
time.
time.
Example
# statistics admin level only for
backend
stats
stats admin if
Example
# statistics admin level always enabled because of the
backend
stats
stats auth
stats admin if
Example
# statistics admin level depends on the authenticated
userlist
group admin users
user admin insecure-password
group readonly users
user haproxy insecure-password
backend
stats
acl AUTH
acl AUTH_ADMIN http_auth_group(stats-auth)
stats http-request auth unless
stats admin if
See also : "stats enable", "stats auth", "stats http-request",
"bind-process", section 3.4 about userlists and section 7
ACL
stats auth
Enable statistics with authentication and grant access to an
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<user> is a user name to grant access
<passwd> is the cleartext password associated to this
This statement enables statistics with default settings, and restricts
to declared users only. It may be repeated as many times as necessary
allow as many users as desired. When a user tries to access the
without a valid account, a "401 Forbidden" response will be returned so
the browser asks the user to provide a valid user and password. The
which will be returned to the browser is configurable using "stats
Since the authentication method is HTTP Basic Authentication, the
circulate in cleartext on the network. Thus, it was decided that
configuration file would also use cleartext passwords to remind the
that those ones should not be sensitive and not shared with any other
It is also possible to reduce the scope of the proxies which appear in
report using "stats
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious
Example
# public access (limited to this backend
backend
server srv1 192.168.0.
stats
stats
stats
stats uri
stats realm Haproxy\
stats auth
stats auth
# internal monitoring access
backend
stats
stats uri
stats refresh
See also : "stats enable", "stats realm", "stats scope", "stats
stats
Enable statistics reporting with default
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
This statement enables statistics reporting with default settings
at build time. Unless stated otherwise, these settings are used
- stats uri :
- stats realm : "HAProxy
- stats auth : no
- stats scope : no
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious
Example
# public access (limited to this backend
backend
server srv1 192.168.0.
stats
stats
stats
stats uri
stats realm Haproxy\
stats auth
stats auth
# internal monitoring access
backend
stats
stats uri
stats refresh
See also : "stats auth", "stats realm", "stats
stats
Enable statistics and hide HAProxy version
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
By default, the stats page reports some useful status information along
the statistics. Among them is HAProxy's version. However, it is
considered dangerous to report precise version to anyone, as it can help
target known weaknesses with specific attacks. The "stats
statement removes the version from the statistics report. This is
for public sites or any site with a weak
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious
Example
# public access (limited to this backend
backend
server srv1 192.168.0.
stats
stats
stats
stats uri
stats realm Haproxy\
stats auth
stats auth
# internal monitoring access
backend
stats
stats uri
stats refresh
See also : "stats auth", "stats enable", "stats realm", "stats
stats http-request { allow | deny | auth [realm <realm>]
[ { if | unless } <condition>
Access control for
May be used in sections: defaults | frontend | listen |
no | no | yes |
As "http-request", these set of options allow to fine control access
statistics. Each option may be followed by if/unless and
First option with matched condition (or option without condition) is
For "deny" a 403 error will be returned, for "allow" normal processing
performed, for "auth" a 401/407 error code is returned so the
should be asked to enter a username and
There is no fixed limit to the number of http-request statements
instance.
instance.
See also : "http-request", section 3.4 about userlists and section
about ACL
stats realm
Enable statistics and set authentication
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<realm> is the name of the HTTP Basic Authentication realm reported
the browser. The browser uses it to display it in the
inviting the user to enter a valid username and
The realm is read as a single word, so any spaces in it should be
using a backslash
This statement is useful only in conjunction with "stats auth" since it
only related to
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious
Example
# public access (limited to this backend
backend
server srv1 192.168.0.
stats
stats
stats
stats uri
stats realm Haproxy\
stats auth
stats auth
# internal monitoring access
backend
stats
stats uri
stats refresh
See also : "stats auth", "stats enable", "stats
stats refresh
Enable statistics with automatic
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<delay> is the suggested refresh delay, specified in seconds, which
be returned to the browser consulting the report page. While
browser is free to apply any delay, it will generally respect
and refresh the page this every seconds. The refresh interval
be specified in any other non-default time unit, by suffixing
unit after the value, as explained at the top of this
This statement is useful on monitoring displays with a permanent
reporting the load balancer's activity. When set, the HTML report page
include a link "refresh"/"stop refresh" so that the user can select
he wants automatic refresh of the page or
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious
Example
# public access (limited to this backend
backend
server srv1 192.168.0.
stats
stats
stats
stats uri
stats realm Haproxy\
stats auth
stats auth
# internal monitoring access
backend
stats
stats uri
stats refresh
See also : "stats auth", "stats enable", "stats realm", "stats
stats scope { <name> | "."
Enable statistics and limit access
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<name> is the name of a listen, frontend or backend section to
reported. The special name "." (a single dot) designates
section in which the statement
When this statement is specified, only the sections enumerated with
statement will appear in the report. All other ones will be hidden.
statement may appear as many times as needed if multiple sections need to
reported. Please note that the name checking is performed as simple
comparisons, and that it is never checked that a give section name
exists.
exists.
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious
Example
# public access (limited to this backend
backend
server srv1 192.168.0.
stats
stats
stats
stats uri
stats realm Haproxy\
stats auth
stats auth
# internal monitoring access
backend
stats
stats uri
stats refresh
See also : "stats auth", "stats enable", "stats realm", "stats
stats show-desc [ <description>
Enable reporting of a description on the statistics
May be used in sections : defaults | frontend | listen |
yes | no | yes |
<name> is an optional description to be reported. If unspecified,
description from global section is automatically used
This statement is useful for users that offer shared services to
customers, where node or description should be different for each
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious parameters. By default description is not
Example
# internal monitoring access
backend
stats
stats show-desc Master node for Europe, Asia,
stats uri
stats refresh
See also: "show-node", "stats enable", "stats uri" and "description"
global
stats
Enable reporting additional informations on the statistics page
- cap: capabilities
- mode: one of tcp, http or health
- id: SNMP ID (proxy, socket,
- IP (socket,
- cookie (backend,
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious parameters. Default behaviour is not to show this
See also: "stats enable", "stats
stats show-node [ <name>
Enable reporting of a host name on the statistics
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments:
Arguments:
<name> is an optional name to be reported. If unspecified,
node name from global section is automatically used
This statement is useful for users that offer shared services to
customers, where node or description might be different on a stats
provided for each customer. Default behaviour is not to show host
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious
Example:
Example:
# internal monitoring access
backend
stats
stats show-node
stats uri
stats refresh
See also: "show-desc", "stats enable", "stats uri", and "node" in
section.
section.
stats uri
Enable statistics and define the URI prefix to access
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<prefix> is the prefix of any URI which will be redirected to stats.
prefix may contain a question mark ('?') to indicate part of
query
The statistics URI is intercepted on the relayed traffic, so it appears as
page within the normal application. It is strongly advised to ensure that
selected URI will never appear in the application, otherwise it will never
possible to reach it in the
The default URI compiled in haproxy is "/haproxy?stats", but this may
changed at build time, so it's better to always explicitly specify it
It is generally a good idea to include a question mark in the URI so
intermediate proxies refrain from caching the results. Also, since any
beginning with the prefix will be accepted as a stats request, the
mark helps ensuring that no valid URI will begin with the same
It is sometimes very convenient to use "/" as the URI prefix, and put
statement in a "listen" instance of its own. That makes it easy to
an address or a port to statistics
Though this statement alone is enough to enable statistics reporting, it
recommended to set all other settings in order to avoid relying on
unobvious
Example
# public access (limited to this backend
backend
server srv1 192.168.0.
stats
stats
stats
stats uri
stats realm Haproxy\
stats auth
stats auth
# internal monitoring access
backend
stats
stats uri
stats refresh
See also : "stats auth", "stats enable", "stats
stick match <pattern> [table <table>] [{if | unless}
Define a request pattern matching condition to stick a user to a
May be used in sections : defaults | frontend | listen |
no | no | yes |
Arguments
<pattern> is a pattern extraction rule as described in section 7.8.
describes what elements of the incoming request or
will be analysed in the hope to find a matching entry in
stickiness table. This rule is
<table> is an optional stickiness table name. If unspecified, the
backend's table is used. A stickiness table is declared
the "stick-table"
<cond> is an optional matching condition. It makes it possible to
on a certain criterion only when other conditions are met
not met). For instance, it could be used to match on a source
address except when a request passes through a known proxy,
which case we'd match on a header containing that IP
Some protocols or applications require complex stickiness rules and
always simply rely on cookies nor hashing. The "stick match"
describes a rule to extract the stickiness criterion from an incoming
or connection. See section 7 for a complete list of possible patterns
transformation
The table has to be declared using the "stick-table" statement. It must be
a type compatible with the pattern. By default it is the one which is
in the same backend. It is possible to share a table with other backends
referencing it using the "table" keyword. If another table is
the server's ID inside the backends are used. By default, all server
start at 1 in each backend, so the server ordering is enough. But in case
doubt, it is highly recommended to force server IDs using their "id"
It is possible to restrict the conditions where a "stick match"
will apply, using "if" or "unless" followed by a condition. See section 7
ACL based
There is no limit on the number of "stick match" statements. The first
applies and matches will cause the request to be directed to the same
as was used for the request which created the entry. That way,
matches can be used as
The stick rules are checked after the persistence cookies, so they will
affect stickiness if a cookie has already been used to select a server.
way, it becomes very easy to insert cookies and match on IP addresses
order to maintain stickiness between HTTP and
Note : Consider not using this feature in multi-process mode (nbproc >
unless you know what you do : memory is not shared between
processes, which can result in random
Example
# forward SMTP users to the same server they just used for POP in
# last 30
backend
mode
balance
stick store-request
stick-table type ip size 200k expire
server s1 192.168.1.
server s2 192.168.1.
backend
mode
balance
stick match src table
server s1 192.168.1.
server s2 192.168.1.
See also : "stick-table", "stick on", "nbproc", "bind-process" and section
about ACLs and pattern
stick on <pattern> [table <table>] [{if | unless}
Define a request pattern to associate a user to a
May be used in sections : defaults | frontend | listen |
no | no | yes |
Note : This form is exactly equivalent to "stick match" followed
"stick store-request", all with the same arguments. Please
to both keywords for details. It is only provided as a
for writing more maintainable
Note : Consider not using this feature in multi-process mode (nbproc >
unless you know what you do : memory is not shared between
processes, which can result in random
Examples
# The following
stick on src table pop if
# ...is strictly equivalent to this one
stick match src table pop if
stick store-request src table pop if
# Use cookie persistence for HTTP, and stick on source address for HTTPS
# well as HTTP without cookie. Share the same table between both
backend
mode
balance
stick on src table
cookie SRV insert indirect
server s1 192.168.1.1:80 cookie
server s2 192.168.1.1:80 cookie
backend
mode
balance
stick-table type ip size 200k expire
stick on
server s1 192.168.1.
server s2 192.168.1.
See also : "stick match", "stick store-request", "nbproc" and
stick store-request <pattern> [table <table>] [{if | unless}
Define a request pattern used to create an entry in a stickiness
May be used in sections : defaults | frontend | listen |
no | no | yes |
Arguments
<pattern> is a pattern extraction rule as described in section 7.8.
describes what elements of the incoming request or
will be analysed, extracted and stored in the table once
server is
<table> is an optional stickiness table name. If unspecified, the
backend's table is used. A stickiness table is declared
the "stick-table"
<cond> is an optional storage condition. It makes it possible to
certain criteria only when some conditions are met (or not
For instance, it could be used to store the source IP
except when the request passes through a known proxy, in
case we'd store a converted form of a header containing that
address.
address.
Some protocols or applications require complex stickiness rules and
always simply rely on cookies nor hashing. The "stick store-request"
describes a rule to decide what to extract from the request and when to
it, in order to store it into a stickiness table for further requests
match it using the "stick match" statement. Obviously the extracted part
make sense and have a chance to be matched in a further request. Storing
client's IP address for instance often makes sense. Storing an ID found in
URL parameter also makes sense. Storing a source port will almost never
any sense because it will be randomly matched. See section 7 for a
list of possible patterns and transformation
The table has to be declared using the "stick-table" statement. It must be
a type compatible with the pattern. By default it is the one which is
in the same backend. It is possible to share a table with other backends
referencing it using the "table" keyword. If another table is
the server's ID inside the backends are used. By default, all server
start at 1 in each backend, so the server ordering is enough. But in case
doubt, it is highly recommended to force server IDs using their "id"
It is possible to restrict the conditions where a "stick
statement will apply, using "if" or "unless" followed by a condition.
condition will be evaluated while parsing the request, so any criteria can
used. See section 7 for ACL based
There is no limit on the number of "stick store-request" statements,
there is a limit of 8 simultaneous stores per request or response.
makes it possible to store up to 8 criteria, all extracted from either
request or the response, regardless of the number of rules. Only the 8
ones which match will be kept. Using this, it is possible to feed
tables at once in the hope to increase the chance to recognize a user
another protocol or access
The "store-request" rules are evaluated once the server connection has
established, so that the table will contain the real server that
the
Note : Consider not using this feature in multi-process mode (nbproc >
unless you know what you do : memory is not shared between
processes, which can result in random
Example
# forward SMTP users to the same server they just used for POP in
# last 30
backend
mode
balance
stick store-request
stick-table type ip size 200k expire
server s1 192.168.1.
server s2 192.168.1.
backend
mode
balance
stick match src table
server s1 192.168.1.
server s2 192.168.1.
See also : "stick-table", "stick on", "nbproc", "bind-process" and section
about ACLs and pattern
stick-table type {ip | integer | string [len <length>] } size
[expire <expire>]
Configure the stickiness table for the current
May be used in sections : defaults | frontend | listen |
no | no | yes |
Arguments
ip a table declared with "type ip" will only store IPv4
This form is very compact (about 50 bytes per entry) and
very fast entry lookup and stores with almost no overhead.
is mainly used to store client source IP
integer a table declared with "type integer" will store 32bit
which can represent a client identifier found in a request
instance.
instance.
string a table declared with "type string" will store substrings of
to <len> characters. If the string provided by the
extractor is larger than <len>, it will be truncated
being stored. During matching, at most <len> characters will
compared between the string in the table and the
pattern. When not specified, the string is automatically
to 31
<length> is the maximum number of characters that will be stored in
"string" type table. See type "string" above. Be careful
changing this parameter as memory usage will
increase.
increase.
<size> is the maximum number of entries that can fit in the table.
value directly impacts memory usage. Count
50 bytes per entry, plus the size of a string if any. The
supports suffixes "k", "m", "g" for 2^10, 2^20 and 2^30
[nopurge] indicates that we refuse to purge older entries when the
is full. When not specified and the table is full when
wants to store an entry in it, it will flush a few of the
entries in order to release some space for the new ones. This
most often the desired behaviour. In some specific cases,
be desirable to refuse new entries instead of purging the
ones. That may be the case when the amount of data to store
far above the hardware limits and we prefer not to offer
to new clients than to reject the ones already connected.
using this parameter, be sure to properly set the
parameter (see
<expire> defines the maximum duration of an entry in the table since
was last created, refreshed or matched. The expiration delay
defined using the standard time format, similarly as the
timeouts. The maximum duration is slightly above 24 days.
section 2.2 for more information. If this delay is not
the session won't automatically expire, but older entries
be removed once full. Be sure not to use the "nopurge"
if not expiration delay is
The is only one stick-table per backend. At the moment of writing this
it does not seem useful to have multiple tables per backend. If this
to be required, simply create a dummy backend with a stick-table in it
reference
It is important to understand that stickiness based on learning
has some limitations, including the fact that all learned associations
lost upon restart. In general it can be good as a complement but not
as an exclusive
See also : "stick match", "stick on", "stick store-request", and section 2.
about time
tcp-request content accept [{if | unless}
Accept a connection if/unless a content inspection condition is
May be used in sections : defaults | frontend | listen |
no | yes | yes |
During TCP content inspection, the connection is immediately validated if
condition is true (when used with "if") or false (when used with
Most of the time during content inspection, a condition will be in
uncertain state which is neither true nor false. The evaluation
stops when such a condition is encountered. It is important to
that "accept" and "reject" rules are evaluated in their exact
order, so that it is possible to build complex rules from them. There is
specific limit to the number of rules which may be
Note that the "if/unless" condition is optional. If no condition is set
the action, it is simply performed
If no "tcp-request content" rules are matched, the default action already
"accept". Thus, this statement alone does not bring anything without
"reject"
See section 7 about ACL
See also : "tcp-request content reject", "tcp-request
tcp-request content reject [{if | unless}
Reject a connection if/unless a content inspection condition is
May be used in sections : defaults | frontend | listen |
no | yes | yes |
During TCP content inspection, the connection is immediately rejected if
condition is true (when used with "if") or false (when used with
Most of the time during content inspection, a condition will be in
uncertain state which is neither true nor false. The evaluation
stops when such a condition is encountered. It is important to
that "accept" and "reject" rules are evaluated in their exact
order, so that it is possible to build complex rules from them. There is
specific limit to the number of rules which may be
Note that the "if/unless" condition is optional. If no condition is set
the action, it is simply performed
If no "tcp-request content" rules are matched, the default action is set
"accept".
Example:
"accept".
Example:
# reject SMTP connection if client speaks
tcp-request inspect-delay
acl content_present req_len gt
tcp-request reject if
# Forward HTTPS connection only if client
tcp-request inspect-delay
acl content_present req_len gt
tcp-request accept if
tcp-request
See section 7 about ACL
See also : "tcp-request content accept", "tcp-request
tcp-request inspect-delay
Set the maximum allowed time to wait for data during content
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<timeout> is the timeout value specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
People using haproxy primarily as a TCP relay are often worried about
risk of passing any type of protocol to a server without any analysis.
order to be able to analyze the request contents, we must first
the data then analyze them. This statement simply enables withholding
data for at most the specified amount of
Note that when performing content inspection, haproxy will evaluate the
rules for every new chunk which gets in, taking into account the fact
those data are partial. If no rule matches before the aforementioned
a last check is performed upon expiration, this time considering that
contents are definitive. If no delay is set, haproxy will not wait at
and will immediately apply a verdict based on the available
Obviously this is unlikely to be very useful and might even be racy, so
setups are not
As soon as a rule matches, the request is released and continues as usual.
the timeout is reached and no rule matches, the default policy will be to
it pass through
For most protocols, it is enough to set it to a few seconds, as most
send the full request immediately upon connection. Add 3 or more seconds
cover TCP retransmits but that's all. For some protocols, it may make
to use large values, for instance to ensure that the client never
before the server (eg: SMTP), or to wait for a client to talk before
data to the server (eg: SSL). Note that the client timeout must cover
least the inspection delay, otherwise it will expire first. If the
closes the connection or if the buffer is full, the delay immediately
since the contents will not be able to change
See also : "tcp-request content accept", "tcp-request content
"timeout
timeout check
Set additional check timeout, but only after a connection has been
established.
established.
May be used in sections: defaults | frontend | listen |
yes | no | yes |
Arguments:
Arguments:
<timeout> is the timeout value specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
If set, haproxy uses min("timeout connect", "inter") as a connect
for check and "timeout check" as an additional read timeout. The "min"
used so that people running with *very* long "timeout connect" (eg.
who needed this due to the queue or tarpit) do not slow down their
(Please also note that there is no valid reason to have such long
timeouts, because "timeout queue" and "timeout tarpit" can always be used
avoid
If "timeout check" is not set haproxy uses "inter" for complete
timeout (connect + read) exactly like all <1.3.15
In most cases check request is much simpler and faster to handle than
requests and people may want to kick out laggy servers so this timeout
be smaller than "timeout
This parameter is specific to backends, but can be specified once for all
"defaults" sections. This is in fact one of the easiest solutions not
forget about
See also: "timeout connect", "timeout queue", "timeout
"timeout
timeout client
timeout clitimeout <timeout>
Set the maximum inactivity time on the client
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<timeout> is the timeout value specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
The inactivity timeout applies when the client is expected to acknowledge
send data. In HTTP mode, this timeout is particularly important to
during the first phase, when the client sends the request, and during
response while it is reading data sent by the server. The value is
in milliseconds by default, but can be in any other unit if the number
suffixed by the unit, as specified at the top of this document. In TCP
(and to a lesser extent, in HTTP mode), it is highly recommended that
client timeout remains equal to the server timeout in order to avoid
situations to debug. It is a good practice to cover one or several TCP
losses by specifying timeouts that are slightly above multiples of 3
(eg: 4 or 5
This parameter is specific to frontends, but can be specified once for all
"defaults" sections. This is in fact one of the easiest solutions not
forget about it. An unspecified timeout results in an infinite timeout,
is not recommended. Such a usage is accepted and works but reports a
during startup because it may results in accumulation of expired sessions
the system if the system's timeouts are not configured
This parameter replaces the old, deprecated "clitimeout". It is
to use it to write new configurations. The form "timeout clitimeout"
provided only by backwards compatibility but its use is strongly
See also : "clitimeout", "timeout
timeout connect
timeout contimeout <timeout>
Set the maximum time to wait for a connection attempt to a server to
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<timeout> is the timeout value specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
If the server is located on the same LAN as haproxy, the connection should
immediate (less than a few milliseconds). Anyway, it is a good practice
cover one or several TCP packet losses by specifying timeouts that
slightly above multiples of 3 seconds (eg: 4 or 5 seconds). By default,
connect timeout also presets both queue and tarpit timeouts to the same
if these have not been
This parameter is specific to backends, but can be specified once for all
"defaults" sections. This is in fact one of the easiest solutions not
forget about it. An unspecified timeout results in an infinite timeout,
is not recommended. Such a usage is accepted and works but reports a
during startup because it may results in accumulation of failed sessions
the system if the system's timeouts are not configured
This parameter replaces the old, deprecated "contimeout". It is
to use it to write new configurations. The form "timeout contimeout"
provided only by backwards compatibility but its use is strongly
See also: "timeout check", "timeout queue", "timeout server",
"timeout
timeout http-keep-alive
Set the maximum allowed time to wait for a new HTTP request to
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<timeout> is the timeout value specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
By default, the time to wait for a new request in case of keep-alive is
by "timeout http-request". However this is not always convenient because
people want very short keep-alive timeouts in order to release
faster, and others prefer to have larger ones but still have short
once the request has started to present
The "http-keep-alive" timeout covers these needs. It will define how long
wait for a new HTTP request to start coming after a response was sent.
the first byte of request has been seen, the "http-request" timeout is
to wait for the complete request to come. Note that empty lines prior to
new request do not refresh the timeout and are not counted as a new
There is also another difference between the two timeouts : when a
expires during timeout http-keep-alive, no error is returned, the
just closes. If the connection expires in "http-request" while waiting for
connection to complete, a HTTP 408 error is
In general it is optimal to set this value to a few tens to hundreds
milliseconds, to allow users to fetch all objects of a page at once
without waiting for further clicks. Also, if set to a very small value
1 millisecond) it will probably only accept pipelined requests but not
non-pipelined ones. It may be a nice trade-off for very large sites
with tens to hundreds of thousands of
If this parameter is not set, the "http-request" timeout applies, and if
are not set, "timeout client" still applies at the lower level. It should
set in the frontend to take effect, unless the frontend is in TCP mode,
which case the HTTP backend's timeout will be
See also : "timeout http-request", "timeout
timeout http-request
Set the maximum allowed time to wait for a complete HTTP
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<timeout> is the timeout value specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
In order to offer DoS protection, it may be required to lower the
accepted time to receive a complete HTTP request without affecting the
timeout. This helps protecting against established connections on
nothing is sent. The client timeout cannot offer a good protection
this abuse because it is an inactivity timeout, which means that if
attacker sends one character every now and then, the timeout will
trigger. With the HTTP request timeout, no matter what speed the
types, the request will be aborted if it does not complete in
Note that this timeout only applies to the header part of the request,
not to any data. As soon as the empty line is received, this timeout is
used anymore. It is used again on keep-alive connections to wait for a
request if "timeout http-keep-alive" is not
Generally it is enough to set it to a few seconds, as most clients send
full request immediately upon connection. Add 3 or more seconds to cover
retransmits but that's all. Setting it to very low values (eg: 50 ms)
generally work on local networks as long as there are no packet losses.
will prevent people from sending bare HTTP requests using
If this parameter is not set, the client timeout still applies between
chunk of the incoming request. It should be set in the frontend to
effect, unless the frontend is in TCP mode, in which case the HTTP
timeout will be
See also : "timeout http-keep-alive", "timeout
timeout queue
Set the maximum time to wait in the queue for a connection slot to be
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<timeout> is the timeout value specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
When a server's maxconn is reached, connections are left pending in a
which may be server-specific or global to the backend. In order not to
indefinitely, a timeout is applied to requests pending in the queue. If
timeout is reached, it is considered that the request will almost never
served, so it is dropped and a 503 error is returned to the
The "timeout queue" statement allows to fix the maximum time for a request
be left pending in a queue. If unspecified, the same value as the
connection timeout ("timeout connect") is used, for backwards
with older versions with no "timeout queue"
See also : "timeout connect",
timeout server
timeout srvtimeout <timeout>
Set the maximum inactivity time on the server
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments
<timeout> is the timeout value specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
The inactivity timeout applies when the server is expected to acknowledge
send data. In HTTP mode, this timeout is particularly important to
during the first phase of the server's response, when it has to send
headers, as it directly represents the server's processing time for
request. To find out what value to put there, it's often good to start
what would be considered as unacceptable response times, then check the
to observe the response time distribution, and adjust the value
The value is specified in milliseconds by default, but can be in any
unit if the number is suffixed by the unit, as specified at the top of
document. In TCP mode (and to a lesser extent, in HTTP mode), it is
recommended that the client timeout remains equal to the server timeout
order to avoid complex situations to debug. Whatever the expected
response times, it is a good practice to cover at least one or several
packet losses by specifying timeouts that are slightly above multiples of
seconds (eg: 4 or 5 seconds
This parameter is specific to backends, but can be specified once for all
"defaults" sections. This is in fact one of the easiest solutions not
forget about it. An unspecified timeout results in an infinite timeout,
is not recommended. Such a usage is accepted and works but reports a
during startup because it may results in accumulation of expired sessions
the system if the system's timeouts are not configured
This parameter replaces the old, deprecated "srvtimeout". It is
to use it to write new configurations. The form "timeout srvtimeout"
provided only by backwards compatibility but its use is strongly
See also : "srvtimeout", "timeout
timeout tarpit
Set the duration for which tarpitted connections will be
May be used in sections : defaults | frontend | listen |
yes | yes | yes |
Arguments
<timeout> is the tarpit duration specified in milliseconds by default,
can be in any other unit if the number is suffixed by the
as explained at the top of this
When a connection is tarpitted using "reqtarpit", it is maintained open
no activity for a certain amount of time, then closed. "timeout
defines how long it will be maintained
The value is specified in milliseconds by default, but can be in any
unit if the number is suffixed by the unit, as specified at the top of
document. If unspecified, the same value as the backend's connection
("timeout connect") is used, for backwards compatibility with older
with no "timeout tarpit"
See also : "timeout connect",
transparent
Enable client-side transparent
May be used in sections : defaults | frontend | listen |
yes | no | yes |
Arguments :
This keyword was introduced in order to provide layer 7 persistence to
3 load balancers. The idea is to use the OS's ability to redirect an
connection for a remote address to a local process (here HAProxy), and
this process know what address was initially requested. When this option
used, sessions without cookies will be forwarded to the original
IP address of the incoming request (which should match that of
equipment), while requests with cookies will still be forwarded to
appropriate
The "transparent" keyword is deprecated, use "option transparent"
Note that contrary to a common belief, this option does NOT make
present the client's IP to the server when establishing the
See also: "option
use_backend <backend> if
use_backend <backend> unless
Switch to a specific backend if/unless an ACL-based condition is
May be used in sections : defaults | frontend | listen |
no | yes | yes |
Arguments
<backend> is the name of a valid backend or "listen"
<condition> is a condition composed of ACLs, as described in section
When doing content-switching, connections arrive on a frontend and are
dispatched to various backends depending on a number of conditions.
relation between the conditions and the backends is described with
"use_backend" keyword. While it is normally used with HTTP processing, it
also be used in pure TCP, either without content using stateless ACLs
source address validation) or combined with a "tcp-request" rule to wait
some
There may be as many "use_backend" rules as desired. All of these rules
evaluated in their declaration order, and the first one which matches
assign the
In the first form, the backend will be used if the condition is met. In
second form, the backend will be used if the condition is not met. If
condition is valid, the backend defined with "default_backend" will be
If no default backend is defined, either the servers in the same section
used (in case of a "listen" section) or, in case of a frontend, no server
used and a 503 service unavailable response is
Note that it is possible to switch from a TCP frontend to an HTTP backend.
this case, either the frontend has already checked that the protocol is
and backend processing will immediately follow, or the backend will wait
a complete HTTP request to get in. This feature is useful when a
must decode several protocols on a unique port, one of them being
See also: "default_backend", "tcp-request", and section 7 about
5. Server and default-server
------------------------------------
------------------------------------
The "server" and "default-server" keywords support a certain number of
which are all passed as arguments on the server line. The order in which
arguments appear does not count, and they are all optional. Some of
settings are single words (booleans) while others expect one or several
after them. In this case, the values must immediately follow the setting
Except default-server, all those settings must be specified after the
address if they are
server <name> <address>[:port] [settings ...
default-server [settings ...
The currently supported settings are the following
addr
Using the "addr" parameter, it becomes possible to use a different IP
to send health-checks. On some servers, it may be desirable to dedicate an
address to specific component able to perform complex tests which are
suitable to health-checks than the application. This parameter is ignored
the "check" parameter is not set. See also the "port"
Supported in default-server:
backup
backup
When "backup" is present on a server line, the server is only used in
balancing when all other non-backup servers are unavailable. Requests
with a persistence cookie referencing the server will always be
though. By default, only the first operational backup server is used,
the "allbackups" option is set in the backend. See also the
option.
option.
Supported in default-server:
check
check
This option enables health checks on the server. By default, a server
always considered available. If "check" is set, the server will
periodic health checks to ensure that it is really able to serve
The default address and port to send the tests to are those of the
and the default source is the same as the one defined in the backend. It
possible to change the address using the "addr" parameter, the port using
"port" parameter, the source address using the "source" address, and
interval and timers using the "inter", "rise" and "fall" parameters.
request method is define in the backend using the "httpchk",
"mysql-check" and "ssl-hello-chk" options. Please refer to those options
parameters for more
Supported in default-server:
cookie
The "cookie" parameter sets the cookie value assigned to the server
<value>. This value will be checked in incoming requests, and the
operational server possessing the same value will be selected. In return,
cookie insertion or rewrite modes, this value will be assigned to the
sent to the client. There is nothing wrong in having several servers
the same cookie value, and it is in fact somewhat common between normal
backup servers. See also the "cookie" keyword in backend
Supported in default-server:
disabled
disabled
The "disabled" keyword starts the server in the "disabled" state. That
that it is marked down in maintenance mode, and no connection other than
ones allowed by persist mode will reach it. It is very well suited to
new servers, because normal traffic will never reach them, while it is
possible to test the service by making use of the force-persist
Supported in default-server:
error-limit
If health observing is enabled, the "error-limit" parameter specifies
number of consecutive errors that triggers event selected by the
option. By default it is set to 10 consecutive
Supported in default-server:
See also the "check", "error-limit" and
fall
The "fall" parameter states that a server will be considered as dead
<count> consecutive unsuccessful health checks. This value defaults to 3
unspecified. See also the "check", "inter" and "rise"
Supported in default-server:
id
Set a persistent ID for the server. This ID must be positive and unique
the proxy. An unused ID will automatically be assigned if unset. The
assigned value will be 1. This ID is currently only returned in
Supported in default-server:
inter
fastinter
downinter
The "inter" parameter sets the interval between two consecutive health
to <delay> milliseconds. If left unspecified, the delay defaults to 2000
It is also possible to use "fastinter" and "downinter" to optimize
between checks depending on the server state
Server state | Interval
---------------------------------+-----------------------------------------
---------------------------------+-----------------------------------------
UP 100% (non-transitional) |
---------------------------------+-----------------------------------------
---------------------------------+-----------------------------------------
Transitionally UP (going down),
Transitionally DOWN (going up), | "fastinter" if set, "inter"
or yet unchecked.
---------------------------------+-----------------------------------------
---------------------------------+-----------------------------------------
DOWN 100% (non-transitional) | "downinter" if set, "inter"
---------------------------------+-----------------------------------------
---------------------------------+-----------------------------------------
Just as with every other time-based parameter, they can be entered in
other explicit unit among { us, ms, s, m, h, d }. The "inter" parameter
serves as a timeout for health checks sent to servers if "timeout check"
not set. In order to reduce "resonance" effects when multiple servers
hosted on the same hardware, the health-checks of all servers are
with a small time offset between them. It is also possible to add some
noise in the health checks interval using the global
keyword. This makes sense for instance when a lot of backends use the
servers.
servers.
Supported in default-server:
maxconn
The "maxconn" parameter specifies the maximal number of
connections that will be sent to this server. If the number of
concurrent requests goes higher than this value, they will be queued,
for a connection to be released. This parameter is very important as it
save fragile servers from going down under extreme loads. If a
parameter is specified, the limit becomes dynamic. The default value is
which means unlimited. See also the "minconn" and "maxqueue" parameters,
the backend's "fullconn"
Supported in default-server:
maxqueue
The "maxqueue" parameter specifies the maximal number of connections
will wait in the queue for this server. If this limit is reached,
requests will be redispatched to other servers instead of
waiting to be served. This will break persistence but may allow people
quickly re-log in when the server they try to connect to is dying.
default value is "0" which means the queue is unlimited. See also
"maxconn" and "minconn"
Supported in default-server:
minconn
When the "minconn" parameter is set, the maxconn limit becomes a
limit following the backend's load. The server will always accept at
<minconn> connections, never more than <maxconn>, and the limit will be
the ramp between both values when the backend has less than
concurrent connections. This makes it possible to limit the load on
server during normal loads, but push it further for important loads
overloading the server during exceptional loads. See also the
and "maxqueue" parameters, as well as the "fullconn" backend
Supported in default-server:
observe
This option enables health adjusting based on observing communication
the server. By default this functionality is disabled and enabling it
requires to enable health checks. There are two supported modes: "layer4"
"layer7". In layer4 mode, only successful/unsuccessful tcp connections
significant. In layer7, which is only allowed for http proxies,
received from server are verified, like valid/wrong http code,
headers, a timeout, etc. Valid status codes include 100 to 499, 501 and
Supported in default-server:
See also the "check", "on-error" and
on-error
Select what should happen when enough consecutive errors are
Currently, four modes are
- fastinter: force
- fail-check: simulate a failed check, also forces fastinter
- sudden-death: simulate a pre-fatal failed health check, one more
check will mark a server down, forces
- mark-down: mark the server immediately down and force
Supported in default-server:
See also the "check", "observe" and
port
Using the "port" parameter, it becomes possible to use a different port
send health-checks. On some servers, it may be desirable to dedicate a
to a specific component able to perform complex tests which are more
to health-checks than the application. It is common to run a simple script
inetd for instance. This parameter is ignored if the "check" parameter is
set. See also the "addr"
Supported in default-server:
redir
The "redir" parameter enables the redirection mode for all GET and
requests addressing this server. This means that instead of having
forward the request to the server, it will send an "HTTP 302" response
the "Location" header composed of this prefix immediately followed by
requested URI beginning at the leading '/' of the path component. That
that no trailing slash should be used after <prefix>. All invalid
will be rejected, and all non-GET or HEAD requests will be normally served
the server. Note that since the response is completely forged, no
mangling nor cookie insertion is possible in the response. However, cookies
requests are still analysed, making this solution completely usable to
users to a remote location in case of local disaster. Main use consists
increasing bandwidth for static servers by having the clients
connect to them. Note: never use a relative location here, it would cause
loop between the client and
Example : server srv1 192.168.1.1:80 redir http://image1.mydomain.com
Supported in default-server:
rise
The "rise" parameter states that a server will be considered as
after <count> consecutive successful health checks. This value defaults to
if unspecified. See also the "check", "inter" and "fall"
Supported in default-server:
slowstart
The "slowstart" parameter for a server accepts a value in milliseconds
indicates after how long a server which has just come back up will run
full speed. Just as with every other time-based parameter, it can be
in any other explicit unit among { us, ms, s, m, h, d }. The speed
linearly from 0 to 100% during this time. The limitation applies to
parameters
- maxconn: the number of connections accepted by the server will grow from
to 100% of the usual dynamic limit defined by
- weight: when the backend uses a dynamic weighted algorithm, the
grows linearly from 1 to 100%. In this case, the weight is updated at
health-check. For this reason, it is important that the "inter"
is smaller than the "slowstart", in order to maximize the number of
The slowstart never applies when haproxy starts, otherwise it would
trouble to running servers. It only applies when a server has been
seen as
Supported in default-server:
source <addr>[:<pl>[-<ph>]] [usesrc { <addr2>[:<port2>] | client | clientip }
source <addr>[:<port>] [usesrc { <addr2>[:<port2>] | hdr_ip(<hdr>[,<occ>]) }
source <addr>[:<pl>[-<ph>]] [interface
The "source" parameter sets the source address which will be used
connecting to the server. It follows the exact same parameters and
as the backend "source" keyword, except that it only applies to the
referencing it. Please consult the "source" keyword for
Additionally, the "source" statement on a server line allows one to specify
source port range by indicating the lower and higher bounds delimited by
dash ('-'). Some operating systems might require a valid IP address when
source port range is specified. It is permitted to have the same IP/range
several servers. Doing so makes it possible to bypass the maximum of
total concurrent connections. The limit will then reach 64k connections
server.
server.
Supported in default-server:
track
This option enables ability to set the current state of the server
tracking another one. Only a server with checks enabled can be
so it is not possible for example to track a server that tracks
one. If <proxy> is omitted the current one is used. If disable-on-404
used, it has to be enabled on both
Supported in default-server:
weight
The "weight" parameter is used to adjust the server's weight relative
other servers. All servers will receive a load proportional to their
relative to the sum of all weights, so the higher the weight, the higher
load. The default weight is 1, and the maximal value is 256. A value of
means the server will not participate in load-balancing but will still
persistent connections. If this parameter is used to distribute the
according to server's capacity, it is recommended to start with values
can both grow and shrink, for instance between 10 and 100 to leave
room above and below for later
Supported in default-server:
6. HTTP header
---------------------------
---------------------------
In HTTP mode, it is possible to rewrite, add or delete some of the request
response headers based on regular expressions. It is also possible to block
request or a response if a particular header matches a regular
which is enough to stop most elementary protocol attacks, and to
against information leak from the internal network. But there is a
to this : since HAProxy's HTTP engine does not support keep-alive, only
passed during the first request of a TCP session will be seen. All
headers will be considered data only and not analyzed. Furthermore,
never touches data contents, it stops analysis at the end of
There is an exception though. If HAProxy encounters an "Informational
(status code 1xx), it is able to process all rsp* rules which can allow,
rewrite or delete a header, but it will refuse to add a header to any
messages as this is not HTTP-compliant. The reason for still processing
in such responses is to stop and/or fix any possible information leak which
happen, for instance because another downstream equipment would
add a header, or if a server name appears there. When such messages are
normal processing still occurs on the next non-informational
This section covers common usage of the following keywords, described in
in section 4.2
- reqadd
- reqallow
- reqiallow
- reqdel
- reqidel
- reqdeny
- reqideny
- reqpass
- reqipass
- reqrep <search>
- reqirep <search>
- reqtarpit
- reqitarpit
- rspadd
- rspdel
- rspidel
- rspdeny
- rspideny
- rsprep <search>
- rspirep <search>
With all these keywords, the same conventions are used. The <search>
is a POSIX extended regular expression (regex) which supports grouping
parenthesis (without the backslash). Spaces and other delimiters must
prefixed with a backslash ('\') to avoid confusion with a field
Other characters may be prefixed with a backslash to change their meaning
\t for a
\r for a carriage return
\n for a new line
\ to mark a space and differentiate it from a
\# to mark a sharp and differentiate it from a
\\ to use a backslash in a
\\\\ to use a backslash in the text (*2 for regex, *2 for
\xXX to write the ASCII hex code XX as in the C
The <replace> parameter contains the string to be used to replace the
portion of text matching the regex. It can make use of the special
above, and can reference a substring which is delimited by parenthesis in
regex, by writing a backslash ('\') immediately followed by one digit from 0
9 indicating the group position (0 designating the entire line). This
is very common to users of the "sed"
The <string> parameter represents the string which will systematically be
after the last header line. It can also use special character sequences
Notes related to these keywords
---------------------------------
---------------------------------
- these keywords are not always convenient to allow/deny based on
contents. It is strongly recommended to use ACLs with the "block"
instead, resulting in far more flexible and manageable
- lines are always considered as a whole. It is not possible to
a header name only or a value only. This is important because of the
headers are written (notably the number of spaces after the
- the first line is always considered as a header, which makes it possible
rewrite or filter HTTP requests URIs or response codes, but in turn
it harder to distinguish between headers and request line. The regex
^[^\ \t]*[\ \t] matches any HTTP method followed by a space, and the
^[^ \t:]*: matches any header name followed by a
- for performances reasons, the number of characters added to a request or
a response is limited at build time to values between 1 and 4 kB.
should normally be far more than enough for most usages. If it is too
on occasional usages, it is possible to gain some space by removing
useless headers before adding new
- keywords beginning with "reqi" and "rspi" are the same as their
without the 'i' letter except that they ignore case when matching
- when a request passes through a frontend then a backend, all req*
from the frontend will be evaluated, then all req* rules from the
will be evaluated. The reverse path is applied to
- req* statements are applied after "block" statements, so that "block"
always the first one, but before "use_backend" in order to permit
before
7. Using ACLs and pattern
------------------------------------
------------------------------------
The use of Access Control Lists (ACL) provides a flexible solution to
content switching and generally to take decisions based on content
from the request, the response or any environmental status. The principle
simple
- define test criteria with sets of
- perform actions only if a set of tests is
The actions generally consist in blocking the request, or selecting a
In order to define a test, the "acl" keyword is used. The syntax is
acl <aclname> <criterion> [flags] [operator]
This creates a new ACL <aclname> or completes an existing one with new
Those tests apply to the portion of request/response specified in
and may be adjusted with optional flags [flags]. Some criteria also
an operator which may be specified before the set of values. The values
of the type supported by the criterion, and are separated by
ACL names must be formed from upper and lower case letters, digits, '-'
'_' (underscore) , '.' (dot) and ':' (colon). ACL names are
which means that "my_acl" and "My_Acl" are two different
There is no enforced limit to the number of ACLs. The unused ones do not
performance, they just consume a small amount of
The following ACL flags are currently supported
-i : ignore case during matching of all subsequent
-f : load patterns from a
-- : force end of flags. Useful when a string looks like one of the
The "-f" flag is special as it loads all of the lines it finds in the
specified in argument and loads all of them before continuing. It is
possible to pass multiple "-f" arguments if the patterns are to be loaded
multiple files. Empty lines as well as lines beginning with a sharp ('#')
be ignored. All leading spaces and tabs will be stripped. If it is
needed to insert a valid pattern beginning with a sharp, just prefix it with
space so that it is not taken for a comment. Depending on the data type
match method, haproxy may load the lines into a binary tree, allowing very
lookups. This is true for IPv4 and exact string matching. In this
duplicates will automatically be removed. Also, note that the "-i" flag
to subsequent entries and not to entries loaded from files preceeding it.
instance
acl valid-ua hdr(user-agent) -f exact-ua.lst -i -f generic-ua.lst
In this example, each line of "exact-ua.lst" will be exactly matched
the "user-agent" header of the request. Then each line of "generic-ua" will
case-insensitively matched. Then the word "test" will be insensitively
too.
too.
Note that right now it is difficult for the ACL parsers to report errors, so
a file is unreadable or unparsable, the most you'll get is a parse error in
ACL. Thus, file-based ACLs should only be produced by reliable
Supported types of values are
- integers or integer
-
- regular
- IP addresses and
7.1. Matching
----------------------
----------------------
Matching integers is special in that ranges and operators are permitted.
that integer matching only applies to positive values. A range is a
expressed with a lower and an upper bound separated with a colon, both of
may be
For instance, "1024:65535" is a valid range to represent a range
unprivileged ports, and "1024:" would also work. "0:1023" is a
representation of privileged ports, and ":1023" would also
As a special case, some ACL functions support decimal numbers which are in
two integers separated by a dot. This is used with some version checks
instance. All integer properties apply to those decimal numbers,
ranges and
For an easier usage, comparison operators are also supported. Note that
operators with ranges does not make much sense and is strongly
Similarly, it does not make much sense to perform order comparisons with a
of
Available operators for integer matching are
eq : true if the tested value equals at least one
ge : true if the tested value is greater than or equal to at least one
gt : true if the tested value is greater than at least one
le : true if the tested value is less than or equal to at least one
lt : true if the tested value is less than at least one
For instance, the following ACL matches any negative Content-Length header
acl negative-length hdr_val(content-length) lt
This one matches SSL versions between 3.0 and 3.1 (inclusive)
acl sslv3 req_ssl_ver 3:3.
7.2. Matching
---------------------
---------------------
String matching applies to verbatim strings as they are passed, with
exception of the backslash ("\") which makes it possible to escape
characters such as the space. If the "-i" flag is passed before the
string, then the matching will be performed ignoring the case. In
to match the string "-i", either set it second, or pass the "--"
before the first string. Same applies of course to match the string
7.3. Matching regular expressions
-------------------------------------------
-------------------------------------------
Just like with string matching, regex matching applies to verbatim strings
they are passed, with the exception of the backslash ("\") which makes
possible to escape some characters such as the space. If the "-i" flag
passed before the first regex, then the matching will be performed
the case. In order to match the string "-i", either set it second, or
the "--" flag before the first string. Same principle applies of course
match the string
7.4. Matching IPv4
----------------------------
----------------------------
IPv4 addresses values can be specified either as plain addresses or with
netmask appended, in which case the IPv4 address matches whenever it
within the network. Plain addresses may also be replaced with a
host name, but this practice is generally discouraged as it makes it
difficult to read and debug configurations. If hostnames are used, you
at least ensure that they are present in /etc/hosts so that the
does not depend on any random DNS match at the moment the configuration
parsed.
parsed.
7.5. Available matching
--------------------------------
--------------------------------
7.5.1. Matching at Layer 4 and
------------------------------------
------------------------------------
A first set of criteria applies to information which does not require
analysis of the request or response contents. Those generally include
addresses and ports, as well as internal values independant on the
always_false
always_false
This one never matches. All values and flags are ignored. It may be used
a temporary replacement for another one when adjusting
always_true
always_true
This one always matches. All values and flags are ignored. It may be used
a temporary replacement for another one when adjusting
avg_queue
avg_queue(backend)
Returns the total number of queued connections of the designated
divided by the number of active servers. This is very similar to
except that the size of the farm is considered, in order to give a
accurate measurement of the time it may take for a new connection to
processed. The main usage is to return a sorry page to new users when
becomes certain they will get a degraded service. Note that in the
there would not be any active server anymore, we would consider twice
number of queued connections as the measured value. This is a fair
as we expect one server to get back soon anyway, but we still prefer to
new traffic to another backend if in better shape. See also the
"be_conn", and "be_sess_rate"
be_conn
be_conn(backend)
Applies to the number of currently established connections on the
possibly including the connection being evaluated. If no backend name
specified, the current one is used. But it is also possible to check
backend. It can be used to use a specific farm when the nominal one is
See also the "fe_conn", "queue" and "be_sess_rate"
be_id
Applies to the backend's id. Can be used in frontends to check from
backend it was
be_sess_rate
be_sess_rate(backend)
Returns true when the sessions creation rate on the backend matches
specified values or ranges, in number of new sessions per second. This
used to switch to an alternate backend when an expensive or fragile
reaches too high a session rate, or to limit abuse of service (eg.
sucking of an online
Example
# Redirect to an error page if the dictionary is requested too
backend
mode
acl being_scanned be_sess_rate gt
redirect location /denied.html if
connslots
connslots(backend)
The basic idea here is to be able to measure the number of connection
still available (connection + queue), so that anything beyond that
usage; see "use_backend" keyword) can be redirected to a different
'connslots' = number of available server connection slots, + number
available server queue
Note that while "fe_conn" may be used, "connslots" comes in
useful when you have a case of traffic going to one single ip, splitting
multiple backends (perhaps using acls to do name-based load balancing)
you want to be able to differentiate between different backends, and
available "connslots". Also, whereas "nbsrv" only measures servers that
actually *down*, this acl is more fine-grained and looks into the number
available connection slots as well. See also "queue" and
OTHER CAVEATS AND NOTES: at this point in time, the code does not take
of dynamic connections. Also, if any of the server maxconn, or maxqueue is
then this acl clearly does not make sense, in which case the value
will be
dst
Applies to the local IPv4 address the client connected to. It can be used
switch to a different backend for some alternative
dst_conn
Applies to the number of currently established connections on the same
including the one being evaluated. It can be used to either return a
page before hard-blocking, or to use a specific backend to drain new
when the socket is considered saturated. This offers the ability to
different limits to different listening ports or addresses. See also
"fe_conn" and "be_conn"
dst_port
Applies to the local port the client connected to. It can be used to
to a different backend for some alternative
fe_conn
fe_conn(frontend)
Applies to the number of currently established connections on the
possibly including the connection being evaluated. If no frontend name
specified, the current one is used. But it is also possible to check
frontend. It can be used to either return a sorry page before
or to use a specific backend to drain new requests when the farm
considered saturated. See also the "dst_conn", "be_conn" and
criteria.
criteria.
fe_id
Applies to the frontend's id. Can be used in backends to check from
frontend it was
fe_sess_rate
fe_sess_rate(frontend)
Returns true when the session creation rate on the current or the
frontend matches the specified values or ranges, expressed in new
per second. This is used to limit the connection rate to acceptable ranges
order to prevent abuse of service at the earliest moment. This can
combined with layer 4 ACLs in order to force the clients to wait a bit
the rate to go down below the
Example
# This frontend limits incoming mails to 10/s with a max of
# concurrent connections. We accept any connection below 10/s,
# force excess clients to wait for 100 ms. Since clients are limited
# 100 max, there cannot be more than 10 incoming mails per
frontend
bind
mode
maxconn
acl too_fast fe_sess_rate ge
tcp-request inspect-delay
tcp-request content accept if !
tcp-request content accept if
nbsrv
nbsrv(backend)
Returns true when the number of usable servers of either the current
or the named backend matches the values or ranges specified. This is used
switch to an alternate backend when the number of servers is too low
to handle some load. It is useful to report a failure when combined
"monitor
queue
queue(backend)
Returns the total number of queued connections of the designated
including all the connections in server queues. If no backend name
specified, the current one is used, but it is also possible to check
one. This can be used to take actions when queuing goes above a known
generally indicating a surge of traffic or a massive slowdown on the
One possible action could be to reject new users but still accept old
See also the "avg_queue", "be_conn", and "be_sess_rate"
so_id
Applies to the socket's id. Useful in frontends with many bind
src
Applies to the client's IPv4 address. It is usually used to limit access
certain resources such as statistics. Note that it is the TCP-level
address which is used, and not the address of a client behind a
src_port
Applies to the client's TCP source port. This has a very limited
srv_id
Applies to the server's id. Can be used in frontends or
srv_is_up(<server>)
srv_is_up(<backend>/<server>)
srv_is_up(<server>)
srv_is_up(<backend>/<server>)
Returns true when the designated server is UP, and false when it is
DOWN or in maintenance mode. If <backend> is omitted, then the server
looked up in the current backend. The function takes no arguments since
is used as a boolean. It is mainly used to take action based on an
status reported via a health check (eg: a geographical site's
Another possible use which is more of a hack consists in using dummy
as boolean variables that can be enabled or disabled from the CLI, so
rules depending on those ACLs can be tweaked in
7.5.2. Matching contents at Layer
-----------------------------------
-----------------------------------
A second set of criteria depends on data found in buffers, but which can
during analysis. This requires that some data has been buffered, for
through TCP request content inspection. Please see the "tcp-request"
for more detailed information on the
req_len
Returns true when the length of the data in the request buffer matches
specified range. It is important to understand that this test does
return false as long as the buffer is changing. This means that a check
equality to zero will almost always immediately match at the beginning of
session, while a test for more data will wait for that data to come in
return false only when haproxy is certain that no more data will come
This test was designed to be used with TCP request content
req_proto_http
req_proto_http
Returns true when data in the request buffer look like HTTP and
parses as such. It is the same parser as the common HTTP request parser
is used so there should be no surprises. This test can be used for
to direct HTTP traffic to a given port and HTTPS traffic to another
using TCP request content inspection
req_rdp_cookie
req_rdp_cookie(name)
Returns true when data in the request buffer look like the RDP protocol,
a cookie is present and equal to <string>. By default, any cookie name
checked, but a specific cookie name can be specified in parenthesis.
parser only checks for the first cookie, as illustrated in the RDP
specification. The cookie name is case insensitive. This ACL can be
with the "MSTS" cookie, as it can contain the user name of the
connecting to the server if properly configured on the client. This can
used to restrict access to certain servers to certain
req_rdp_cookie_cnt
req_rdp_cookie_cnt(name)
Returns true when the data in the request buffer look like the RDP
and the number of RDP cookies matches the specified range (typically zero
one). Optionally a specific cookie name can be checked. This is a simple
of detecting the RDP protocol, as clients generally send the MSTS or
cookies.
cookies.
req_ssl_ver
Returns true when data in the request buffer look like SSL, with a
version matching the specified range. Both SSLv2 hello messages and
messages are supported. The test tries to be strict enough to avoid
easily fooled. In particular, it waits for as many bytes as announced in
message header if this header looks valid (bound to the buffer size).
that TLSv1 is announced as SSL version 3.1. This test was designed to be
with TCP request content
wait_end
wait_end
Waits for the end of the analysis period to return true. This may be used
conjunction with content analysis to avoid returning a wrong verdict
It may also be used to delay some actions, such as a delayed reject for
special addresses. Since it either stops the rules evaluation or
returns true, it is recommended to use this acl as the last one in a
Please note that the default ACL "WAIT_END" is always usable without
declaration. This test was designed to be used with TCP request
inspection.
inspection.
Examples
# delay every incoming request by 2
tcp-request inspect-delay
tcp-request content accept if
# don't immediately tell bad guys they are
tcp-request inspect-delay
acl goodguys src 10.0.0.
acl badguys src 10.0.1.
tcp-request content accept if
tcp-request content reject if badguys
tcp-request content
7.5.3. Matching at Layer
--------------------------
--------------------------
A third set of criteria applies to information which can be found at
application layer (layer 7). Those require that a full HTTP request has
read, and are only evaluated then. They may require slightly more CPU
than the layer 4 ones, but not much since the request and response are
hdr
hdr(header)
Note: all the "hdr*" matching criteria either apply to all headers, or to
particular header whose name is passed between parenthesis and without
space. The header name is not case-sensitive. The header matching
with RFC2616, and treats as separate headers all values delimited by
Use the shdr() variant for response headers sent by the
The "hdr" criteria returns true if any of the headers matching the
match any of the strings. This can be used to check exact for values.
instance, checking that "connection: close" is set
hdr(Connection) -i
hdr_beg
hdr_beg(header)
Returns true when one of the headers begins with one of the strings.
"hdr" for more information on header matching. Use the shdr_beg() variant
response headers sent by the
hdr_cnt
hdr_cnt(header)
Returns true when the number of occurrence of the specified header
the values or ranges specified. It is important to remember that one
line may count as several headers if it has several values. This is used
detect presence, absence or abuse of a specific header, as well as to
request smuggling attacks by rejecting requests which contain more than
of certain headers. See "hdr" for more information on header matching.
the shdr_cnt() variant for response headers sent by the
hdr_dir
hdr_dir(header)
Returns true when one of the headers contains one of the strings
isolated or delimited by slashes. This is used to perform filename
directory name matching, and may be used with Referer. See "hdr" for
information on header matching. Use the shdr_dir() variant for
headers sent by the
hdr_dom
hdr_dom(header)
Returns true when one of the headers contains one of the strings
isolated or delimited by dots. This is used to perform domain name
and may be used with the Host header. See "hdr" for more information
header matching. Use the shdr_dom() variant for response headers sent by
server.
server.
hdr_end
hdr_end(header)
Returns true when one of the headers ends with one of the strings. See
for more information on header matching. Use the shdr_end() variant
response headers sent by the
hdr_ip
hdr_ip(header)
Returns true when one of the headers' values contains an IP address
<ip_address>. This is mainly used with headers such as X-Forwarded-For
X-Client-IP. See "hdr" for more information on header matching. Use
shdr_ip() variant for response headers sent by the
hdr_len
hdr_len(<header>)
Returns true when at least one of the headers has a length which matches
values or ranges specified. This may be used to detect empty or too
headers. See "hdr" for more information on header matching. Use
shdr_len() variant for response headers sent by the
hdr_reg
hdr_reg(header)
Returns true when one of the headers matches of the regular expressions.
can be used at any time, but it is important to remember that regex
is slower than other methods. See also other "hdr_" criteria, as well
"hdr" for more information on header matching. Use the shdr_reg() variant
response headers sent by the
hdr_sub
hdr_sub(header)
Returns true when one of the headers contains one of the strings. See
for more information on header matching. Use the shdr_sub() variant
response headers sent by the
hdr_val
hdr_val(header)
Returns true when one of the headers starts with a number which matches
values or ranges specified. This may be used to limit content-length
acceptable values for example. See "hdr" for more information on
matching. Use the shdr_val() variant for response headers sent by the
http_auth(userlist)
http_auth(userlist)
http_auth_group(userlist) <group>
Returns true when authentication data received from the client
username & password stored on the userlist. It is also possible
use http_auth_group to check if the user is assigned to at least
of specified
Currently only http basic auth is
http_first_req
http_first_req
Returns true when the request being processed is the first one of
connection. This can be used to add or remove headers that may be
from some requests when a request is not the first one, or even to
some specific ACL checks only on the first
method
Applies to the method in the HTTP request, eg: "GET". Some predefined
already check for most common
path
Returns true when the path part of the request, which starts at the
slash and ends before the question mark, equals one of the strings. It may
used to match known files, such as /favicon.
path_beg
Returns true when the path begins with one of the strings. This can be
to send certain directory names to alternative
path_dir
Returns true when one of the strings is found isolated or delimited
slashes in the path. This is used to perform filename or directory
matching without the risk of wrong match due to colliding prefixes. See
"url_dir" and
path_dom
Returns true when one of the strings is found isolated or delimited with
in the path. This may be used to perform domain name matching in
requests. See also "path_sub" and
path_end
Returns true when the path ends with one of the strings. This may be used
control file name
path_len
Returns true when the path length matches the values or ranges
This may be used to detect abusive requests for
path_reg
Returns true when the path matches one of the regular expressions. It can
used any time, but it is important to remember that regex matching is
than other methods. See also "url_reg" and all "path_"
path_sub
Returns true when the path contains one of the strings. It can be used
detect particular patterns in paths, such as "../" for example. See
"path_dir".
"path_dir".
req_ver
Applies to the version string in the HTTP request, eg: "1.0". Some
ACL already check for versions 1.0 and 1.
status
Applies to the HTTP status code in the HTTP response, eg: "302". It can
used to act on responses depending on status ranges, for instance,
any Location header if the response is not a
url
Applies to the whole URL passed in the request. The only real use is to
"*", for which there already is a predefined
url_beg
Returns true when the URL begins with one of the strings. This can be used
check whether a URL begins with a slash or with a protocol
url_dir
Returns true when one of the strings is found isolated or delimited
slashes in the URL. This is used to perform filename or directory
matching without the risk of wrong match due to colliding prefixes. See
"path_dir" and
url_dom
Returns true when one of the strings is found isolated or delimited with
in the URL. This is used to perform domain name matching without the risk
wrong match due to colliding prefixes. See also
url_end
Returns true when the URL ends with one of the strings. It has very
use. "path_end" should be used instead for filename
url_ip
Applies to the IP address specified in the absolute URI in an HTTP
It can be used to prevent access to certain resources such as local
It is useful with option
url_len
Returns true when the url length matches the values or ranges specified.
may be used to detect abusive requests for
url_port
Applies to the port specified in the absolute URI in an HTTP request. It
be used to prevent access to certain resources. It is useful with
"http_proxy". Note that if the port is not specified in the request, port
is
url_reg
Returns true when the URL matches one of the regular expressions. It can
used any time, but it is important to remember that regex matching is
than other methods. See also "path_reg" and all "url_"
url_sub
Returns true when the URL contains one of the strings. It can be used
detect particular patterns in query strings for example. See also
7.6. Pre-defined
---------------------
---------------------
Some predefined ACLs are hard-coded so that they do not have to be declared
every frontend which needs them. They all have their names in upper case
order to avoid confusion. Their equivalence is provided
ACL name Equivalent to
---------------+-----------------------------+---------------------------------
---------------+-----------------------------+---------------------------------
FALSE always_false never
HTTP req_proto_http match if protocol is valid
HTTP_1.0 req_ver 1.0 match HTTP version 1.
HTTP_1.1 req_ver 1.1 match HTTP version 1.
HTTP_CONTENT hdr_val(content-length) gt 0 match an existing
HTTP_URL_ABS url_reg ^[^/:]*:// match absolute URL with
HTTP_URL_SLASH url_beg / match URL beginning with
HTTP_URL_STAR url * match URL equal to
LOCALHOST src 127.0.0.1/8 match connection from local
METH_CONNECT method CONNECT match HTTP CONNECT
METH_GET method GET HEAD match HTTP GET or HEAD
METH_HEAD method HEAD match HTTP HEAD
METH_OPTIONS method OPTIONS match HTTP OPTIONS
METH_POST method POST match HTTP POST
METH_TRACE method TRACE match HTTP TRACE
RDP_COOKIE req_rdp_cookie_cnt gt 0 match presence of an RDP
REQ_CONTENT req_len gt 0 match data in the request
TRUE always_true always
WAIT_END wait_end wait for end of content
---------------+-----------------------------+---------------------------------
---------------+-----------------------------+---------------------------------
7.7. Using ACLs to form
----------------------------------
----------------------------------
Some actions are only performed upon a valid condition. A condition is
combination of ACLs with operators. 3 operators are supported
- AND
- OR (explicit with the "or" keyword or the "||"
- Negation with the exclamation mark
A condition is formed as a disjunctive
[!]acl1 [!]acl2 ... [!]acln { or [!]acl1 [!]acl2 ... [!]acln
Such conditions are generally used after an "if" or "unless"
indicating when the condition will trigger the
For instance, to block HTTP requests to the "*" URL with methods other
"OPTIONS", as well as POST requests without content-length, and GET or
requests with a content-length greater than 0, and finally every request
is not either GET/HEAD/POST/OPTIONS
acl missing_cl hdr_cnt(Content-length) eq
block if HTTP_URL_STAR !METH_OPTIONS || METH_POST
block if METH_GET
block unless METH_GET or METH_POST or
To select a different backend for requests to static contents on the "www"
and to every request on the "img", "video", "download" and "ftp" hosts
acl url_static path_beg /static /images /img
acl url_static path_end .gif .png .jpg .css .
acl host_www hdr_beg(host) -i
acl host_static hdr_beg(host) -i img. video. download.
# now use backend "static" for all static-only hosts, and for static
# of host "www". Use backend "www" for the
use_backend static if host_static or host_www
use_backend www if
It is also possible to form rules using "anonymous ACLs". Those are unnamed
expressions that are built on the fly without needing to be declared. They
be enclosed between braces, with a space before and after each brace
the braces must be seen as independant words). Example
The following rule
acl missing_cl hdr_cnt(Content-length) eq
block if METH_POST
Can also be written that way
block if METH_POST { hdr_cnt(Content-length) eq 0
It is generally not recommended to use this construct because it's a lot
to leave errors in the configuration when written that way. However, for
simple rules matching only one source IP address for instance, it can make
sense to use them than to declare ACLs with random names. Another example
good use is the following
With named ACLs
acl site_dead nbsrv(dynamic) lt
acl site_dead nbsrv(static) lt
monitor fail if
With anonymous ACLs
monitor fail if { nbsrv(dynamic) lt 2 } || { nbsrv(static) lt 2
See section 4.2 for detailed help on the "block" and "use_backend"
7.8. Pattern
-----------------------
-----------------------
The stickiness features relies on pattern extraction in the request
response. Sometimes the data needs to be converted first before being
for instance converted from ASCII to IP or upper case to lower
All these operations of data extraction and conversion are defined
"pattern extraction rules". A pattern rule always has the same format.
begins with a single pattern fetch word, potentially followed by a list
arguments within parenthesis then an optional list of transformations.
much as possible, the pattern fetch functions use the same name as
equivalent used in
The list of currently supported pattern fetch functions is the following
src This is the source IPv4 address of the client of the
It is of type IP and only works with such
dst This is the destination IPv4 address of the session on
client side, which is the address the client connected
It can be useful when running in transparent mode. It is
type IP and only works with such
dst_port This is the destination TCP port of the session on the
side, which is the port the client connected to. This might
used when running in transparent mode or when assigning
ports to some clients for a whole application session. It is
type integer and only works with such
hdr(name) This extracts the last occurrence of header <name> in an
request and converts it to an IP address. This IP address
then used to match the table. A typical use is with
x-forwarded-for
The currently available list of transformations include
lower Convert a string pattern to lower case. This can only be
after a string pattern fetch function or after a
function returning a string type. The result is of type
upper Convert a string pattern to upper case. This can only be
after a string pattern fetch function or after a
function returning a string type. The result is of type
ipmask(mask) Apply a mask to an IPv4 address, and use the result for
and storage. This can be used to make all hosts within
certain mask to share the same table entries and as such
the same server. The mask can be passed in dotted form
255.255.255.0) or in CIDR form (eg:
8.
----------
----------
One of HAProxy's strong points certainly lies is its precise logs. It
provides the finest level of information available for such a product, which
very important for troubleshooting complex environments. Standard
provided in logs include client ports, TCP/HTTP state timers, precise
state at termination and precise termination cause, information about
to direct traffic to a server, and of course the ability to capture
headers.
headers.
In order to improve administrators reactivity, it offers a great
about encountered problems, both internal and external, and it is possible
send logs to different sources at the same time with different level filters
- global process-level logs (system errors, start/stop, etc..
- per-instance system and internal errors (lack of resource, bugs, ...
- per-instance external troubles (servers up/down, max
- per-instance activity (client connections), either at the establishment
at the
The ability to distribute different levels of logs to different log
allow several production teams to interact and to fix their problems as
as possible. For example, the system team might monitor system-wide
while the application team might be monitoring the up/down for their servers
real time, and the security team might analyze the activity logs with one
delay.
delay.
8.1. Log
---------------
---------------
TCP and HTTP connections can be logged with information such as the date,
source IP address, destination address, connection duration, response
HTTP request, HTTP return code, number of bytes transmitted,
in which the session ended, and even exchanged cookies values. For
track a particular user's problems. All messages may be sent to up to
syslog servers. Check the "log" keyword in section 4.2 for more
about log
8.2. Log
----------------
----------------
HAProxy supports 4 log formats. Several fields are common between these
and will be detailed in the following sections. A few of them may
slightly with the configuration, due to indicators specific to
options. The supported formats are as follows
- the default format, which is very basic and very rarely used. It
provides very basic information about the incoming connection at the
it is accepted : source IP:port, destination IP:port, and
This mode will eventually disappear so it will not be described to
extents.
extents.
- the TCP format, which is more advanced. This format is enabled when
tcplog" is set on the frontend. HAProxy will then usually wait for
connection to terminate before logging. This format provides much
information, such as timers, connection counts, queue size, etc...
format is recommended for pure TCP
- the HTTP format, which is the most advanced for HTTP proxying. This
is enabled when "option httplog" is set on the frontend. It provides
same information as the TCP format with some HTTP-specific fields such
the request, the status code, and captures of headers and cookies.
format is recommended for HTTP
- the CLF HTTP format, which is equivalent to the HTTP format, but with
fields arranged in the same order as the CLF format. In this mode,
timers, captures, flags, etc... appear one per field after the end of
common fields, in the same order they appear in the standard HTTP
Next sections will go deeper into details for each of these formats.
specification will be performed on a "field" basis. Unless stated otherwise,
field is a portion of text delimited by any number of spaces. Since
servers are susceptible of inserting fields at the beginning of a line, it
always assumed that the first field is the one containing the process name
identifier.
identifier.
Note : Since log lines may be quite long, the log examples in sections
might be broken into multiple lines. The example log lines will
prefixed with 3 closing angle brackets ('>>>') and each time a log
broken into multiple lines, each non-final line will end with
backslash ('\') and the next line will start indented by two
8.2.1. Default log
-------------------------
-------------------------
This format is used when no specific option is set. The log is emitted as
as the connection is accepted. One should note that this currently is the
format which logs the request's destination IP and
Example
listen
mode
log
server srv1 127.0.0.
>>> Feb 6 12:12:09 localhost
haproxy[14385]: Connect from 10.0.1.2:33312 to 10.0.3.31:8012
(www/HTTP)
(www/HTTP)
Field Format Extract from the example
1 process_name '[' pid ']:'
2 'Connect from' Connect
3 source_ip ':' source_port 10.0.1.
4 'to'
5 destination_ip ':' destination_port 10.0.3.
6 '(' frontend_name '/' mode ')'
Detailed fields description
- "source_ip" is the IP address of the client which initiated the
- "source_port" is the TCP port of the client which initiated the
- "destination_ip" is the IP address the client connected
- "destination_port" is the TCP port the client connected
- "frontend_name" is the name of the frontend (or listener) which
and processed the
- "mode is the mode the frontend is operating (TCP or
It is advised not to use this deprecated format for newer installations as
will eventually
8.2.2. TCP log
---------------------
---------------------
The TCP format is used when "option tcplog" is specified in the frontend,
is the recommended format for pure TCP proxies. It provides a lot of
information for troubleshooting. Since this format includes timers and
counts, the log is normally emitted at the end of the session. It can
emitted earlier if "option logasap" is specified, which makes sense in
environments with long sessions such as remote terminals. Sessions which
the "monitor" rules are never logged. It is also possible not to emit logs
sessions for which no data were exchanged between the client and the server,
specifying "option dontlognull" in the frontend. Successful connections
not be logged if "option dontlog-normal" is specified in the frontend. A
fields may slightly vary depending on some configuration options, those
marked with a star ('*') after the field name
Example
frontend
mode
option
log
default_backend
backend
server srv1 127.0.0.
>>> Feb 6 12:12:56 localhost
haproxy[14387]: 10.0.1.2:33313 [06/Feb/2009:12:12:51.443] fnt
bck/srv1 0/0/5007 212 -- 0/0/0/0/3
Field Format Extract from the example
1 process_name '[' pid ']:'
2 client_ip ':' client_port 10.0.1.
3 '[' accept_date ']' [06/Feb/2009:12:12:51.
4 frontend_name
5 backend_name '/' server_name
6 Tw '/' Tc '/' Tt*
7 bytes_read*
8 termination_state
9 actconn '/' feconn '/' beconn '/' srv_conn '/' retries*
10 srv_queue '/' backend_queue
Detailed fields description
- "client_ip" is the IP address of the client which initiated the
connection to
- "client_port" is the TCP port of the client which initiated the
- "accept_date" is the exact date when the connection was received by
(which might be very slightly different from the date observed on
network if there was some queuing in the system's backlog). This is
the same date which may appear in any upstream firewall's
- "frontend_name" is the name of the frontend (or listener) which
and processed the
- "backend_name" is the name of the backend (or listener) which was
to manage the connection to the server. This will be the same as
frontend if no switching rule has been applied, which is common for
applications.
applications.
- "server_name" is the name of the last server to which the connection
sent, which might differ from the first one if there were connection
and a redispatch occurred. Note that this server belongs to the
which processed the request. If the connection was aborted before
a server, "<NOSRV>" is indicated instead of a server
- "Tw" is the total time in milliseconds spent waiting in the various
It can be "-1" if the connection was aborted before reaching the
See "Timers" below for more
- "Tc" is the total time in milliseconds spent waiting for the connection
establish to the final server, including retries. It can be "-1" if
connection was aborted before a connection could be established.
"Timers" below for more
- "Tt" is the total time in milliseconds elapsed between the accept and
last close. It covers all possible processings. There is one exception,
"option logasap" was specified, then the time counting stops at the
the log is emitted. In this case, a '+' sign is prepended before the
indicating that the final one will be larger. See "Timers" below for
details.
details.
- "bytes_read" is the total number of bytes transmitted from the server
the client when the log is emitted. If "option logasap" is specified,
this value will be prefixed with a '+' sign indicating that the final
may be larger. Please note that this value is a 64-bit counter, so
analysis tools must be able to handle it without
- "termination_state" is the condition the session was in when the
ended. This indicates the session state, which side caused the end
session to happen, and for what reason (timeout, error, ...). The
flags should be "--", indicating the session was closed by either end
no data remaining in buffers. See below "Session state at
for more
- "actconn" is the total number of concurrent connections on the process
the session was logged. It it useful to detect when some per-process
limits have been reached. For instance, if actconn is close to 512
multiple connection errors occur, chances are high that the system
the process to use a maximum of 1024 file descriptors and that all of
are used. See section 3 "Global parameters" to find how to tune the
- "feconn" is the total number of concurrent connections on the frontend
the session was logged. It is useful to estimate the amount of
required to sustain high loads, and to detect when the frontend's
has been reached. Most often when this value increases by huge jumps, it
because there is congestion on the backend servers, but sometimes it can
caused by a denial of service
- "beconn" is the total number of concurrent connections handled by
backend when the session was logged. It includes the total number
concurrent connections active on servers as well as the number
connections pending in queues. It is useful to estimate the amount
additional servers needed to support high loads for a given
Most often when this value increases by huge jumps, it is because there
congestion on the backend servers, but sometimes it can be caused by
denial of service
- "srv_conn" is the total number of concurrent connections still active
the server when the session was logged. It can never exceed the
configured "maxconn" parameter. If this value is very often close or
to the server's "maxconn", it means that traffic regulation is involved
lot, meaning that either the server's maxconn value is too low, or
there aren't enough servers to process the load with an optimal
time. When only one of the server's "srv_conn" is high, it usually
that this server has some trouble causing the connections to take longer
be processed than on other
- "retries" is the number of connection retries experienced by this
when trying to connect to the server. It must normally be zero, unless
server is being stopped at the same moment the connection was
Frequent retries generally indicate either a network problem
haproxy and the server, or a misconfigured system backlog on the
preventing new connections from being queued. This field may optionally
prefixed with a '+' sign, indicating that the session has experienced
redispatch after the maximal retry count has been reached on the
server. In this case, the server name appearing in the log is the one
connection was redispatched to, and not the first one, though both
sometimes be the same in case of hashing for instance. So as a general
of thumb, when a '+' is present in front of the retry count, this
should not be attributed to the logged
- "srv_queue" is the total number of requests which were processed
this one in the server queue. It is zero when the request has not
through the server queue. It makes it possible to estimate the
server's response time by dividing the time spent in queue by the number
requests in the queue. It is worth noting that if a session experiences
redispatch and passes through two server queues, their positions will
cumulated. A request should not pass through both the server queue and
backend queue unless a redispatch
- "backend_queue" is the total number of requests which were processed
this one in the backend's global queue. It is zero when the request has
gone through the global queue. It makes it possible to estimate the
queue length, which easily translates into a number of missing servers
divided by a server's "maxconn" parameter. It is worth noting that if
session experiences a redispatch, it may pass twice in the backend's
and then both positions will be cumulated. A request should not
through both the server queue and the backend queue unless a
occurs.
occurs.
8.2.3. HTTP log
----------------------
----------------------
The HTTP format is the most complete and the best suited for HTTP proxies.
is enabled by when "option httplog" is specified in the frontend. It
the same level of information as the TCP format with additional features
are specific to the HTTP protocol. Just like the TCP format, the log is
emitted at the end of the session, unless "option logasap" is specified,
generally only makes sense for download sites. A session which matches
"monitor" rules will never logged. It is also possible not to log sessions
which no data were sent by the client by specifying "option dontlognull" in
frontend. Successful connections will not be logged if "option
is specified in the
Most fields are shared with the TCP log, some being different. A few fields
slightly vary depending on some configuration options. Those ones are
with a star ('*') after the field name
Example
frontend
mode
option
log
default_backend
backend
server srv1 127.0.0.
>>> Feb 6 12:14:14 localhost
haproxy[14389]: 10.0.1.2:33317 [06/Feb/2009:12:14:14.655] http-in
static/srv1 10/0/30/69/109 200 2750 - - ---- 1/1/1/1/0 0/0 {1wt.eu}
{} "GET /index.html HTTP/1.
Field Format Extract from the example
1 process_name '[' pid ']:'
2 client_ip ':' client_port 10.0.1.
3 '[' accept_date ']' [06/Feb/2009:12:14:14.
4 frontend_name
5 backend_name '/' server_name
6 Tq '/' Tw '/' Tc '/' Tr '/' Tt*
7 status_code
8 bytes_read*
9 captured_request_cookie
10 captured_response_cookie
11 termination_state
12 actconn '/' feconn '/' beconn '/' srv_conn '/' retries*
13 srv_queue '/' backend_queue
14 '{' captured_request_headers* '}' {haproxy.1wt.
15 '{' captured_response_headers* '}'
16 '"' http_request '"' "GET /index.html HTTP/1.
Detailed fields description
- "client_ip" is the IP address of the client which initiated the
connection to
- "client_port" is the TCP port of the client which initiated the
- "accept_date" is the exact date when the TCP connection was received
haproxy (which might be very slightly different from the date observed
the network if there was some queuing in the system's backlog). This
usually the same date which may appear in any upstream firewall's log.
does not depend on the fact that the client has sent the request or
- "frontend_name" is the name of the frontend (or listener) which
and processed the
- "backend_name" is the name of the backend (or listener) which was
to manage the connection to the server. This will be the same as
frontend if no switching rule has been
- "server_name" is the name of the last server to which the connection
sent, which might differ from the first one if there were connection
and a redispatch occurred. Note that this server belongs to the
which processed the request. If the request was aborted before reaching
server, "<NOSRV>" is indicated instead of a server name. If the request
intercepted by the stats subsystem, "<STATS>" is indicated
- "Tq" is the total time in milliseconds spent waiting for the client to
a full HTTP request, not counting data. It can be "-1" if the
was aborted before a complete request could be received. It should
be very small because a request generally fits in one single packet.
times here generally indicate network trouble between the client
haproxy. See "Timers" below for more
- "Tw" is the total time in milliseconds spent waiting in the various
It can be "-1" if the connection was aborted before reaching the
See "Timers" below for more
- "Tc" is the total time in milliseconds spent waiting for the connection
establish to the final server, including retries. It can be "-1" if
request was aborted before a connection could be established. See
below for more
- "Tr" is the total time in milliseconds spent waiting for the server to
a full HTTP response, not counting data. It can be "-1" if the request
aborted before a complete response could be received. It generally
the server's processing time for the request, though it may be altered
the amount of data sent by the client to the server. Large times here
"GET" requests generally indicate an overloaded server. See "Timers"
for more
- "Tt" is the total time in milliseconds elapsed between the accept and
last close. It covers all possible processings. There is one exception,
"option logasap" was specified, then the time counting stops at the
the log is emitted. In this case, a '+' sign is prepended before the
indicating that the final one will be larger. See "Timers" below for
details.
details.
- "status_code" is the HTTP status code returned to the client. This
is generally set by the server, but it might also be set by haproxy
the server cannot be reached or when its response is blocked by
- "bytes_read" is the total number of bytes transmitted to the client
the log is emitted. This does include HTTP headers. If "option logasap"
specified, the this value will be prefixed with a '+' sign indicating
the final one may be larger. Please note that this value is a
counter, so log analysis tools must be able to handle it
overflowing.
overflowing.
- "captured_request_cookie" is an optional "name=value" entry indicating
the client had this cookie in the request. The cookie name and its
length are defined by the "capture cookie" statement in the
configuration. The field is a single dash ('-') when the option is
set. Only one cookie may be captured, it is generally used to track
ID exchanges between a client and a server to detect session
between clients due to application bugs. For more details, please
the section "Capturing HTTP headers and cookies"
- "captured_response_cookie" is an optional "name=value" entry
that the server has returned a cookie with its response. The cookie
and its maximum length are defined by the "capture cookie" statement in
frontend configuration. The field is a single dash ('-') when the option
not set. Only one cookie may be captured, it is generally used to
session ID exchanges between a client and a server to detect
crossing between clients due to application bugs. For more details,
consult the section "Capturing HTTP headers and cookies"
- "termination_state" is the condition the session was in when the
ended. This indicates the session state, which side caused the end
session to happen, for what reason (timeout, error, ...), just like in
logs, and information about persistence operations on cookies in the
two characters. The normal flags should begin with "--", indicating
session was closed by either end with no data remaining in buffers.
below "Session state at disconnection" for more
- "actconn" is the total number of concurrent connections on the process
the session was logged. It it useful to detect when some per-process
limits have been reached. For instance, if actconn is close to 512 or
when multiple connection errors occur, chances are high that the
limits the process to use a maximum of 1024 file descriptors and that
of them are used. See section 3 "Global parameters" to find how to tune
system.
system.
- "feconn" is the total number of concurrent connections on the frontend
the session was logged. It is useful to estimate the amount of
required to sustain high loads, and to detect when the frontend's
has been reached. Most often when this value increases by huge jumps, it
because there is congestion on the backend servers, but sometimes it can
caused by a denial of service
- "beconn" is the total number of concurrent connections handled by
backend when the session was logged. It includes the total number
concurrent connections active on servers as well as the number
connections pending in queues. It is useful to estimate the amount
additional servers needed to support high loads for a given
Most often when this value increases by huge jumps, it is because there
congestion on the backend servers, but sometimes it can be caused by
denial of service
- "srv_conn" is the total number of concurrent connections still active
the server when the session was logged. It can never exceed the
configured "maxconn" parameter. If this value is very often close or
to the server's "maxconn", it means that traffic regulation is involved
lot, meaning that either the server's maxconn value is too low, or
there aren't enough servers to process the load with an optimal
time. When only one of the server's "srv_conn" is high, it usually
that this server has some trouble causing the requests to take longer to
processed than on other
- "retries" is the number of connection retries experienced by this
when trying to connect to the server. It must normally be zero, unless
server is being stopped at the same moment the connection was
Frequent retries generally indicate either a network problem
haproxy and the server, or a misconfigured system backlog on the
preventing new connections from being queued. This field may optionally
prefixed with a '+' sign, indicating that the session has experienced
redispatch after the maximal retry count has been reached on the
server. In this case, the server name appearing in the log is the one
connection was redispatched to, and not the first one, though both
sometimes be the same in case of hashing for instance. So as a general
of thumb, when a '+' is present in front of the retry count, this
should not be attributed to the logged
- "srv_queue" is the total number of requests which were processed
this one in the server queue. It is zero when the request has not
through the server queue. It makes it possible to estimate the
server's response time by dividing the time spent in queue by the number
requests in the queue. It is worth noting that if a session experiences
redispatch and passes through two server queues, their positions will
cumulated. A request should not pass through both the server queue and
backend queue unless a redispatch
- "backend_queue" is the total number of requests which were processed
this one in the backend's global queue. It is zero when the request has
gone through the global queue. It makes it possible to estimate the
queue length, which easily translates into a number of missing servers
divided by a server's "maxconn" parameter. It is worth noting that if
session experiences a redispatch, it may pass twice in the backend's
and then both positions will be cumulated. A request should not
through both the server queue and the backend queue unless a
occurs.
occurs.
- "captured_request_headers" is a list of headers captured in the request
to the presence of the "capture request header" statement in the
Multiple headers can be captured, they will be delimited by a vertical
('|'). When no capture is enabled, the braces do not appear, causing
shift of remaining fields. It is important to note that this field
contain spaces, and that using it requires a smarter log parser than
it's not used. Please consult the section "Capturing HTTP headers
cookies" below for more
- "captured_response_headers" is a list of headers captured in the
due to the presence of the "capture response header" statement in
frontend. Multiple headers can be captured, they will be delimited by
vertical bar ('|'). When no capture is enabled, the braces do not
causing a shift of remaining fields. It is important to note that
field may contain spaces, and that using it requires a smarter log
than when it's not used. Please consult the section "Capturing HTTP
and cookies" below for more
- "http_request" is the complete HTTP request line, including the
request and HTTP version string. Non-printable characters are encoded
below the section "Non-printable characters"). This is always the
field, and it is always delimited by quotes and is the only one which
contain quotes. If new fields are added to the log format, they will
added before this field. This field might be truncated if the request
huge and does not fit in the standard syslog buffer (1024 characters).
is the reason why this field must always remain the last
8.3. Advanced logging
-----------------------------
-----------------------------
Some advanced logging options are often looked for but are not easy to find
just by looking at the various options. Here is an entry point for the
options which can enable better logging. Please refer to the keywords
for more information about their
8.3.1. Disabling logging of external
------------------------------------------
------------------------------------------
It is quite common to have some monitoring tools perform health checks
haproxy. Sometimes it will be a layer 3 load-balancer such as LVS or
commercial load-balancer, and sometimes it will simply be a more
monitoring system such as Nagios. When the tests are very frequent, users
ask how to disable logging for those checks. There are three possibilities
- if connections come from everywhere and are just TCP probes, it is
desired to simply disable logging of connections without data exchange,
setting "option dontlognull" in the frontend. It also disables logging
port scans, which may or may not be
- if the connection come from a known source network, use "monitor-net"
declare this network as monitoring only. Any host in this network will
only be able to perform health checks, and their requests will not
logged. This is generally appropriate to designate a list of
such as other
- if the tests are performed on a known URI, use "monitor-uri" to
this URI as dedicated to monitoring. Any host sending this request
only get the result of a health-check, and the request will not be
8.3.2. Logging before waiting for the session to
----------------------------------------------------------
----------------------------------------------------------
The problem with logging at end of connection is that you have no clue
what is happening during very long sessions, such as remote terminal
or large file downloads. This problem can be worked around by
"option logasap" in the frontend. Haproxy will then log as soon as
just before data transfer begins. This means that in case of TCP, it will
log the connection status to the server, and in case of HTTP, it will log
after processing the server headers. In this case, the number of bytes
is the number of header bytes sent to the client. In order to avoid
with normal logs, the total time field and the number of bytes are
with a '+' sign which means that real numbers are certainly
8.3.3. Raising log level upon
------------------------------------
------------------------------------
Sometimes it is more convenient to separate normal traffic from errors
for instance in order to ease error monitoring from log files. When the
"log-separate-errors" is used, connections which experience errors,
retries, redispatches or HTTP status codes 5xx will see their syslog
raised from "info" to "err". This will help a syslog daemon store the log
a separate file. It is very important to keep the errors in the normal
file too, so that log ordering is not altered. You should also be careful
you already have configured your syslog daemon to store all logs higher
"notice" in an "admin" file, because the "err" level is higher than
8.3.4. Disabling logging of successful
--------------------------------------------------
--------------------------------------------------
Although this may sound strange at first, some large sites have to deal
multiple thousands of logs per second and are experiencing difficulties
them intact for a long time or detecting errors within them. If the
"dontlog-normal" is set on the frontend, all normal connections will not
logged. In this regard, a normal connection is defined as one without
error, timeout, retry nor redispatch. In HTTP, the status code is checked
and a response with a status 5xx is not considered normal and will be
too. Of course, doing is is really discouraged as it will remove most of
useful information from the logs. Do this only if you have no
alternative.
alternative.
8.4. Timing
------------------
------------------
Timers provide a great help in troubleshooting network problems. All values
reported in milliseconds (ms). These timers should be used in conjunction
the session termination flags. In TCP mode with "option tcplog" set on
frontend, 3 control points are reported under the form "Tw/Tc/Tt", and in
mode, 5 control points are reported under the form "Tq/Tw/Tc/Tr/Tt"
- Tq: total time to get the client request (HTTP mode only). It's the
elapsed between the moment the client connection was accepted and
moment the proxy received the last HTTP header. The value "-1"
that the end of headers (empty line) has never been seen. This happens
the client closes prematurely or times
- Tw: total time spent in the queues waiting for a connection slot.
accounts for backend queue as well as the server queues, and depends on
queue size, and the time needed for the server to complete
requests. The value "-1" means that the request was killed before
the queue, which is generally what happens with invalid or denied
- Tc: total time to establish the TCP connection to the server. It's the
elapsed between the moment the proxy sent the connection request, and
moment it was acknowledged by the server, or between the TCP SYN packet
the matching SYN/ACK packet in return. The value "-1" means that
connection never
- Tr: server response time (HTTP mode only). It's the time elapsed
the moment the TCP connection was established to the server and the
the server sent its complete response headers. It purely shows its
processing time, without the network overhead due to the data
It is worth noting that when the client has data to send to the server,
instance during a POST request, the time already runs, and this can
apparent response time. For this reason, it's generally wise not to
too much this field for POST requests initiated from clients behind
untrusted network. A value of "-1" here means that the last the
header (empty line) was never seen, most likely because the server
stroke before the server managed to process the
- Tt: total session duration time, between the moment the proxy accepted
and the moment both ends were closed. The exception is when the
option is specified. In this case, it only equals (Tq+Tw+Tc+Tr), and
prefixed with a '+' sign. From this field, we can deduce "Td", the
transmission time, by substracting other timers when valid
Td = Tt - (Tq + Tw + Tc +
Timers with "-1" values have to be excluded from this equation. In
mode, "Tq" and "Tr" have to be excluded too. Note that "Tt" can never
negative.
negative.
These timers provide precious indications on trouble causes. Since the
protocol defines retransmit delays of 3, 6, 12... seconds, we know for
that timers close to multiples of 3s are nearly always related to lost
due to network problems (wires, negotiation, congestion). Moreover, if "Tt"
close to a timeout value specified in the configuration, it often means that
session has been aborted on
Most common cases
- If "Tq" is close to 3000, a packet has probably been lost between
client and the proxy. This is very rare on local networks but might
when clients are on far remote networks and send large requests. It
happen that values larger than usual appear here without any network
Sometimes, during an attack or just after a resource starvation has
haproxy may accept thousands of connections in a few milliseconds. The
spent accepting these connections will inevitably slightly delay
of other connections, and it can happen that request times in the order
a few tens of milliseconds are measured after a few thousands of
connections have been accepted at once. Setting "option
may display larger request times since "Tq" also measures the time
waiting for additional
- If "Tc" is close to 3000, a packet has probably been lost between
server and the proxy during the server connection phase. This value
always be very low, such as 1 ms on local networks and less than a few
of ms on remote
- If "Tr" is nearly always lower than 3000 except some rare values which
to be the average majored by 3000, there are probably some packets
between the proxy and the
- If "Tt" is large even for small byte counts, it generally is
neither the client nor the server decides to close the connection,
instance because both have agreed on a keep-alive connection mode. In
to solve this issue, it will be needed to specify "option httpclose"
either the frontend or the backend. If the problem persists, it means
the server ignores the "close" connection mode and expects the client
close. Then it will be required to use "option forceclose". Having
smallest possible 'Tt' is important when connection regulation is used
the "maxconn" option on the servers, since no new connection will be
to the server until another one is
Other noticeable HTTP log cases ('xx' means any value to be ignored)
Tq/Tw/Tc/Tr/+Tt The "option logasap" is present on the frontend and the
was emitted before the data phase. All the timers are
except "Tt" which is shorter than
-1/xx/xx/xx/Tt The client was not able to send a complete request in
or it aborted too early. Check the session termination
then "timeout http-request" and "timeout client"
Tq/-1/xx/xx/Tt It was not possible to process the request, maybe
servers were out of order, because the request was
or forbidden by ACL rules. Check the session
flags.
flags.
Tq/Tw/-1/xx/Tt The connection could not establish on the server. Either
actively refused it or it timed out after Tt-(Tq+Tw)
Check the session termination flags, then check
"timeout connect" setting. Note that the tarpit action
return similar-looking patterns, with "Tw" equal to the
the client connection was maintained
Tq/Tw/Tc/-1/Tt The server has accepted the connection but did not
a complete response in time, or it closed its
unexpectedly after Tt-(Tq+Tw+Tc) ms. Check the
termination flags, then check the "timeout server"
8.5. Session state at
-----------------------------------
-----------------------------------
TCP and HTTP logs provide a session termination indicator in
"termination_state" field, just before the number of active connections. It
2-characters long in TCP mode, and is extended to 4 characters in HTTP
each of which has a special meaning
- On the first character, a code reporting the first event which caused
session to terminate
C : the TCP session was unexpectedly aborted by the
S : the TCP session was unexpectedly aborted by the server, or
server explicitly refused
P : the session was prematurely aborted by the proxy, because of
connection limit enforcement, because a DENY filter was
because of a security check which detected and blocked a
error in server response which might have caused information
(eg: cacheable cookie), or because the response was processed
the proxy (redirect, stats, etc...
R : a resource on the proxy has been exhausted (memory, sockets,
ports, ...). Usually, this appears during the connection phase,
system logs should contain a copy of the precise error. If
happens, it must be considered as a very serious anomaly
should be fixed as soon as possible by any
I : an internal error was identified by the proxy during a
This should NEVER happen, and you are encouraged to report any
containing this, because this would almost certainly be a bug.
would be wise to preventively restart the process after such
event too, in case it would be caused by memory
c : the client-side timeout expired while waiting for the client
send or receive
s : the server-side timeout expired while waiting for the server
send or receive
- : normal session completion, both the client and the server
with nothing left in the
- on the second character, the TCP or HTTP session state when it was closed
R : the proxy was waiting for a complete, valid REQUEST from the
(HTTP mode only). Nothing was sent to any
Q : the proxy was waiting in the QUEUE for a connection slot. This
only happen when servers have a 'maxconn' parameter set. It
also happen in the global queue after a redispatch consecutive
a failed attempt to connect to a dying server. If no redispatch
reported, then no connection attempt was made to any
C : the proxy was waiting for the CONNECTION to establish on
server. The server might at most have noticed a connection
H : the proxy was waiting for complete, valid response HEADERS from
server (HTTP
D : the session was in the DATA
L : the proxy was still transmitting LAST data to the client while
server had already finished. This one is very rare as it can
happen when the client dies while receiving the last
T : the request was tarpitted. It has been held open with the
during the whole "timeout tarpit" duration or until the
closed, both of which will be reported in the "Tw"
- : normal session completion after end of data
- the third character tells whether the persistence cookie was provided
the client (only in HTTP mode)
N : the client provided NO cookie. This is usually the case for
visitors, so counting the number of occurrences of this flag in
logs generally indicate a valid trend for the site
I : the client provided an INVALID cookie matching no known
This might be caused by a recent configuration change,
cookies between HTTP/HTTPS sites, persistence
ignored, or an
D : the client provided a cookie designating a server which was
so either "option persist" was used and the client was sent
this server, or it was not set and the client was redispatched
another
V : the client provided a VALID cookie, and was sent to the
server.
server.
E : the client provided a valid cookie, but with a last date which
older than what is allowed by the "maxidle" cookie parameter,
the cookie is consider EXPIRED and is ignored. The request will
redispatched just as if there was no
O : the client provided a valid cookie, but with a first date which
older than what is allowed by the "maxlife" cookie parameter,
the cookie is consider too OLD and is ignored. The request will
redispatched just as if there was no
- : does not apply (no cookie set in
- the last character reports what operations were performed on the
cookie returned by the server (only in HTTP mode)
N : NO cookie was provided by the server, and none was inserted
I : no cookie was provided by the server, and the proxy INSERTED
Note that in "cookie insert" mode, if the server provides a
it will still be overwritten and reported as "I"
U : the proxy UPDATED the last date in the cookie that was presented
the client. This can only happen in insert mode with "maxidle".
happens everytime there is activity at a different date than
date indicated in the cookie. If any other change happens, such
a redispatch, then the cookie will be marked as inserted
P : a cookie was PROVIDED by the server and transmitted
R : the cookie provided by the server was REWRITTEN by the proxy,
happens in "cookie rewrite" or "cookie prefix"
D : the cookie provided by the server was DELETED by the
- : does not apply (no cookie set in
The combination of the two first flags gives a lot of information about
was happening when the session terminated, and why it did terminate. It can
helpful to detect server saturation, network troubles, local system
starvation, attacks,
The most common termination flags combinations are indicated below. They
alphabetically sorted, with the lowercase set just after the upper case
easier finding and
Flags
-- Normal
CC The client aborted before the connection could be established to
server. This can happen when haproxy tries to connect to a
dead (or unchecked) server, and the client aborts while haproxy
waiting for the server to respond or for "timeout connect" to
CD The client unexpectedly aborted during data transfer. This can
caused by a browser crash, by an intermediate equipment between
client and haproxy which decided to actively break the
by network routing issues between the client and haproxy, or by
keep-alive session between the server and the client terminated
by the
cD The client did not send nor acknowledge any data for as long as
"timeout client" delay. This is often caused by network failures
the client side, or the client simply leaving the net
CH The client aborted while waiting for the server to start
It might be the server taking too long to respond or the
clicking the 'Stop' button too
cH The "timeout client" stroke while waiting for client data during
POST request. This is sometimes caused by too large TCP MSS
for PPPoE networks which cannot transport full-sized packets. It
also happen when client timeout is smaller than server timeout
the server takes too long to
CQ The client aborted while its session was queued, waiting for a
with enough empty slots to accept it. It might be that either all
servers were saturated or that the assigned server was taking
long a time to
CR The client aborted before sending a full HTTP request. Most
the request was typed by hand using a telnet client, and
too early. The HTTP status code is likely a 400 here. Sometimes
might also be caused by an IDS killing the connection between
and the
cR The "timeout http-request" stroke before the client sent a full
request. This is sometimes caused by too large TCP MSS values on
client side for PPPoE networks which cannot transport
packets, or by clients sending requests by hand and not typing
enough, or forgetting to enter the empty line at the end of
request. The HTTP status code is likely a 408
CT The client aborted while its session was tarpitted. It is important
check if this happens on valid requests, in order to be sure that
wrong tarpit rules have been written. If a lot of them happen,
might make sense to lower the "timeout tarpit" value to
closer to the average reported "Tw" timer, in order not to
resources for just a few
SC The server or an equipment between it and haproxy explicitly
the TCP connection (the proxy received a TCP RST or an ICMP
in return). Under some circumstances, it can also be the
stack telling the proxy that the server is unreachable (eg: no
or no ARP response on local network). When this happens in HTTP
the status code is likely a 502 or 503
sC The "timeout connect" stroke before a connection to the server
complete. When this happens in HTTP mode, the status code is likely
503 or 504
SD The connection to the server died with an error during the
transfer. This usually means that haproxy has received an RST
the server or an ICMP message from an intermediate equipment
exchanging data with the server. This can be caused by a server
or by a network issue on an intermediate
sD The server did not send nor acknowledge any data for as long as
"timeout server" setting during the data phase. This is often
by too short timeouts on L4 equipments before the server
load-balancers, ...), as well as keep-alive sessions
between the client and the server expiring first on
SH The server aborted before sending its full HTTP response headers,
it crashed while processing the request. Since a server aborting
this moment is very rare, it would be wise to inspect its logs
control whether it crashed and why. The logged request may indicate
small set of faulty requests, demonstrating bugs in the
Sometimes this might also be caused by an IDS killing the
between haproxy and the
sH The "timeout server" stroke before the server could return
response headers. This is the most common anomaly, indicating
long transactions, probably caused by server or database
The immediate workaround consists in increasing the "timeout
setting, but it is important to keep in mind that the user
will suffer from these long response times. The only long
solution is to fix the
sQ The session spent too much time in queue and has been expired.
the "timeout queue" and "timeout connect" settings to find out how
fix this if it happens too often. If it often happens massively
short periods, it may indicate general problems on the
servers due to I/O or database congestion, or saturation caused
external
PC The proxy refused to establish a connection to the server because
process' socket limit has been reached while attempting to
The global "maxconn" parameter may be increased in the
so that it does not happen anymore. This status is very rare
might happen when the global "ulimit-n" parameter is forced by
PD The proxy blocked an incorrectly formatted chunked encoded message
a request or a response, after the server has emitted its headers.
most cases, this will indicate an invalid message from the server
the
PH The proxy blocked the server's response, because it was
incomplete, dangerous (cache control), or matched a security
In any case, an HTTP 502 error is sent to the client. One
cause for this error is an invalid syntax in an HTTP header
containing unauthorized characters. It is also possible but
rare, that the proxy blocked a chunked-encoding request from
client due to an invalid syntax, before the server responded. In
case, an HTTP 400 error is sent to the client and reported in
logs.
logs.
PR The proxy blocked the client's HTTP request, either because of
invalid HTTP syntax, in which case it returned an HTTP 400 error
the client, or because a deny filter matched, in which case
returned an HTTP 403
PT The proxy blocked the client's request and has tarpitted
connection before returning it a 500 server error. Nothing was
to the server. The connection was maintained open for as long
reported by the "Tw" timer
RC A local resource has been exhausted (memory, sockets, source
preventing the connection to the server from establishing. The
logs will tell precisely what was missing. This is very rare and
only be solved by proper system
The combination of the two last flags gives a lot of information about
persistence was handled by the client, the server and by haproxy. This is
important to troubleshoot disconnections, when users complain they have
re-authenticate. The commonly encountered flags are
-- Persistence cookie is not
NN No cookie was provided by the client, none was inserted in
response. For instance, this can be in insert mode with
set on a GET
II A cookie designating an invalid server was provided by the
a valid one was inserted in the response. This typically happens
a "server" entry is removed from the configuraton, since its
value can be presented by a client when no other server knows
NI No cookie was provided by the client, one was inserted in
response. This typically happens for first requests from every
in "insert" mode, which makes it an easy way to count real
VN A cookie was provided by the client, none was inserted in
response. This happens for most responses for which the client
already got a
VU A cookie was provided by the client, with a last visit date which
not completely up-to-date, so an updated cookie was provided
response. This can also happen if there was no date at all, or
there was a date but the "maxidle" parameter was not set, so that
cookie can be switched to unlimited
EI A cookie was provided by the client, with a last visit date which
too old for the "maxidle" parameter, so the cookie was ignored and
new cookie was inserted in the
OI A cookie was provided by the client, with a first visit date which
too old for the "maxlife" parameter, so the cookie was ignored and
new cookie was inserted in the
DI The server designated by the cookie was down, a new server
selected and a new cookie was emitted in the
VI The server designated by the cookie was not marked dead but could
be reached. A redispatch happened and selected another one, which
then advertised in the
8.6. Non-printable
-----------------------------
-----------------------------
In order not to cause trouble to log analysis tools or terminals during
consulting, non-printable characters are not sent as-is into log files, but
converted to the two-digits hexadecimal representation of their ASCII
prefixed by the character '#'. The only characters that can be logged
being escaped are comprised between 32 and 126 (inclusive). Obviously,
escape character '#' itself is also encoded to avoid any ambiguity ("#23").
is the same for the character '"' which becomes "#22", as well as '{', '|'
'}' when logging
Note that the space character (' ') is not encoded in headers, which can
issues for tools relying on space count to locate fields. A typical
containing spaces is
Last, it has been observed that some syslog daemons such as syslog-ng
the quote ('"') with a backslash ('\'). The reverse operation can safely
performed since no quote may appear anywhere else in the
8.7. Capturing HTTP
---------------------------
---------------------------
Cookie capture simplifies the tracking a complete user session. This can
achieved using the "capture cookie" statement in the frontend. Please refer
section 4.2 for more details. Only one cookie can be captured, and the
cookie will simultaneously be checked in the request ("Cookie:" header) and
the response ("Set-Cookie:" header). The respective values will be reported
the HTTP logs at the "captured_request_cookie" and
locations (see section 8.2.3 about HTTP log format). When either cookie
not seen, a dash ('-') replaces the value. This way, it's easy to detect when
user switches to a new session for example, because the server will reassign
a new cookie. It is also possible to detect if a server unexpectedly sets
wrong cookie to a client, leading to session
Examples
# capture the first cookie whose name starts with
capture cookie ASPSESSION len
# capture the first cookie whose name is exactly
capture cookie vgnvisitor= len
8.8. Capturing HTTP
---------------------------
---------------------------
Header captures are useful to track unique request identifiers set by an
proxy, virtual host names, user-agents, POST content-length, referrers, etc.
the response, one can search for information about the response length, how
server asked the cache to behave, or an object location during a
Header captures are performed using the "capture request header" and
response header" statements in the frontend. Please consult their definition
section 4.2 for more
It is possible to include both request headers and response headers at the
time. Non-existent headers are logged as empty strings, and if one
appears more than once, only its last occurrence will be logged. Request
are grouped within braces '{' and '}' in the same order as they were
and delimited with a vertical bar '|' without any space. Response
follow the same representation, but are displayed after a space following
request headers block. These blocks are displayed just before the HTTP
in the
Example
# This instance chains to the outgoing
listen
mode
option
option
log
server cache1 192.168.1.
# log the name of the virtual
capture request header Host len
# log the amount of data uploaded during a
capture request header Content-Length len
# log the beginning of the
capture request header Referer len
# server name (useful for outgoing proxies
capture response header Server len
# logging the content-length is useful with "option
capture response header Content-Length len
# log the expected cache behaviour on the
capture response header Cache-Control len
# the Via header will report the next proxy's
capture response header Via len
# log the URL location during a
capture response header Location len
>>> Aug 9 20:26:09 localhost
haproxy[2022]: 127.0.0.1:34014 [09/Aug/2004:20:26:09] proxy-out
proxy-out/cache1 0/0/0/162/+162 200 +350 - - ---- 0/0/0/0/0 0/0
{fr.adserver.yahoo.co||http://fr.f416.mail.} {|864|private||}
"GET http://fr.adserver.yahoo.
>>> Aug 9 20:30:46 localhost
haproxy[2022]: 127.0.0.1:34020 [09/Aug/2004:20:30:46] proxy-out
proxy-out/cache1 0/0/0/182/+182 200 +279 - - ---- 0/0/0/0/0 0/0
{w.ods.org||} {Formilux/0.1.8|3495|||}
"GET http://trafic.1wt.eu/ HTTP/1.
>>> Aug 9 20:30:46 localhost
haproxy[2022]: 127.0.0.1:34028 [09/Aug/2004:20:30:46] proxy-out
proxy-out/cache1 0/0/2/126/+128 301 +223 - - ---- 0/0/0/0/0 0/0
{www.sytadin.equipement.gouv.fr||http://trafic.1wt.eu/}
{Apache|230|||http://www.sytadin.}
"GET http://www.sytadin.equipement.gouv.fr/ HTTP/1.
8.9. Examples of
---------------------
---------------------
These are real-world examples of logs accompanied with an explanation. Some
them have been made up by hand. The syslog part has been removed for
reading. Their sole purpose is to explain how to decipher
>>> haproxy[674]: 127.0.0.1:33318 [15/Oct/2003:08:31:57.130] px-http
px-http/srv1 6559/0/7/147/6723 200 243 - - ---- 5/3/3/1/0 0/0
"HEAD / HTTP/1.
=> long request (6.5s) entered by hand through 'telnet'. The server
in 147 ms, and the session ended normally
>>> haproxy[674]: 127.0.0.1:33319 [15/Oct/2003:08:31:57.149] px-http
px-http/srv1 6559/1230/7/147/6870 200 243 - - ---- 324/239/239/99/0
0/9 "HEAD / HTTP/1.
=> Idem, but the request was queued in the global queue behind 9
requests, and waited there for 1230
>>> haproxy[674]: 127.0.0.1:33320 [15/Oct/2003:08:32:17.654] px-http
px-http/srv1 9/0/7/14/+30 200 +243 - - ---- 3/3/3/1/0 0/0
"GET /image.iso HTTP/1.
=> request for a long data transfer. The "logasap" option was specified,
the log was produced just before transferring data. The server replied
14 ms, 243 bytes of headers were sent to the client, and total time
accept to first data byte is 30
>>> haproxy[674]: 127.0.0.1:33320 [15/Oct/2003:08:32:17.925] px-http
px-http/srv1 9/0/7/14/30 502 243 - - PH-- 3/2/2/0/0 0/0
"GET /cgi-bin/bug.cgi? HTTP/1.
=> the proxy blocked a server response either because of an "rspdeny"
"rspideny" filter, or because the response was improperly formatted
not HTTP-compliant, or because it blocked sensitive information
risked being cached. In this case, the response is replaced with a
bad gateway". The flags ("PH--") tell us that it was haproxy who
to return the 502 and not the
>>> haproxy[18113]: 127.0.0.1:34548 [15/Oct/2003:15:18:55.798] px-http
px-http/<NOSRV> -1/-1/-1/-1/8490 -1 0 - - CR-- 2/2/2/0/0 0/0
=> the client never completed its request and aborted itself ("C---")
8.5s, while the proxy was waiting for the request headers
Nothing was sent to any
>>> haproxy[18113]: 127.0.0.1:34549 [15/Oct/2003:15:19:06.103] px-http
px-http/<NOSRV> -1/-1/-1/-1/50001 408 0 - - cR-- 2/2/2/0/0 0/0
=> The client never completed its request, which was aborted by
time-out ("c---") after 50s, while the proxy was waiting for the
headers ("-R--"). Nothing was sent to any server, but the proxy
send a 408 return code to the
>>> haproxy[18989]: 127.0.0.1:34550 [15/Oct/2003:15:24:28.312] px-tcp
px-tcp/srv1 0/0/5007 0 cD 0/0/0/0/0
=> This log was produced with "option tcplog". The client timed out
5 seconds
>>> haproxy[18989]: 10.0.0.1:34552 [15/Oct/2003:15:26:31.462] px-http
px-http/srv1 3183/-1/-1/-1/11215 503 0 - - SC-- 205/202/202/115/3
0/0 "HEAD / HTTP/1.
=> The request took 3s to complete (probably a network problem), and
connection to the server failed ('SC--') after 4 attempts of 2
(config says 'retries 3'), and no redispatch (otherwise we would
seen "/+3"). Status code 503 was returned to the client. There were
connections on this server, 202 connections on this proxy, and 205
the global process. It is possible that the server refused
connection because of too many already
9. Statistics and
----------------------------
----------------------------
It is possible to query HAProxy about its status. The most commonly
mechanism is the HTTP statistics page. This page also exposes an
CSV output format for monitoring tools. The same format is provided on
Unix
9.1. CSV
---------------
---------------
The statistics may be consulted either from the unix socket or from the
page. Both means provide a CSV format whose fields
0. pxname: proxy
1. svname: service name (FRONTEND for frontend, BACKEND for backend, any
for
2. qcur: current queued
3. qmax: max queued
4. scur: current
5. smax: max
6. slim: sessions
7. stot: total
8. bin: bytes
9. bout: bytes
10. dreq: denied
11. dresp: denied
12. ereq: request
13. econ: connection
14. eresp: response errors (among which
15. wretr: retries
16. wredis: redispatches
17. status: status (UP/DOWN/NOLB/MAINT/MAINT(via)...
18. weight: server weight (server), total weight
19. act: server is active (server), number of active servers
20. bck: server is backup (server), number of backup servers
21. chkfail: number of failed
22. chkdown: number of UP->DOWN
23. lastchg: last status change (in
24. downtime: total downtime (in
25. qlimit: queue
26. pid: process id (0 for first instance, 1 for second, ...
27. iid: unique proxy
28. sid: service id (unique inside a
29. throttle: warm up
30. lbtot: total number of times a server was
31. tracked: id of proxy/server if tracking is
32. type (0=frontend, 1=backend, 2=server,
33. rate: number of sessions per second over last elapsed
34. rate_lim: limit on new sessions per
35. rate_max: max number of new sessions per
36. check_status: status of last health check, one
UNK ->
INI ->
SOCKERR -> socket
L4OK -> check passed on layer 4, no upper layers testing
L4TMOUT -> layer 1-4
L4CON -> layer 1-4 connection problem, for
"Connection refused" (tcp rst) or "No route to host"
L6OK -> check passed on layer
L6TOUT -> layer 6 (SSL)
L6RSP -> layer 6 invalid response - protocol
L7OK -> check passed on layer
L7OKC -> check conditionally passed on layer 7, for example 404
disable-on-404
disable-on-404
L7TOUT -> layer 7 (HTTP/SMTP)
L7RSP -> layer 7 invalid response - protocol
L7STS -> layer 7 response error, for example HTTP
37. check_code: layer5-7 code, if
38. check_duration: time in ms took to finish last health
39. hrsp_1xx: http responses with 1xx
40. hrsp_2xx: http responses with 2xx
41. hrsp_3xx: http responses with 3xx
42. hrsp_4xx: http responses with 4xx
43. hrsp_5xx: http responses with 5xx
44. hrsp_other: http responses with other codes (protocol
45. hanafail: failed health checks
46. req_rate: HTTP requests per second over last elapsed
47. req_rate_max: max number of HTTP requests per second
48. req_tot: total number of HTTP requests
49. cli_abrt: number of data transfers aborted by the
50. srv_abrt: number of data transfers aborted by the server (inc. in
9.2. Unix Socket
-------------------------
-------------------------
The following commands are supported on the UNIX stats socket ; all of
must be terminated by a line feed. The socket supports pipelining, so that
is possible to chain multiple commands at once provided they are delimited
a semi-colon or a line feed, although the former is more reliable as it has
risk of being truncated over the network. The responses themselves will each
followed by an empty line, so it will be easy for an external script to match
given response with a given request. By default one command line is
then the connection closes, but there is an interactive allowing multiple
to be issued one at a
It is important to understand that when multiple haproxy processes are
on the same sockets, any process may pick up the request and will output
own
clear
Clear the max values of the statistics counters in each proxy (frontend
backend) and in each server. The cumulated counters are not affected.
can be used to get clean counters after an incident, without having
restart nor to clear traffic counters. This command is restricted and
only be issued on sockets configured for levels "operator" or
clear counters
Clear all statistics counters in each proxy (frontend & backend) and in
server. This has the same effect as restarting. This command is
and can only be issued on sockets configured for level
disable server
Mark the server DOWN for maintenance. In this mode, no more checks will
performed on the server until it leaves
If the server is tracked by other servers, those servers will be set to
during the
In the statistics page, a server DOWN for maintenance will appear with
"MAINT" status, its tracking servers with the "MAINT(via)"
Both the backend and the server may be specified either by their name or
their numeric ID, prefixed with a sharp
This command is restricted and can only be issued on sockets configured
level
enable server
If the server was previously marked as DOWN for maintenance, this marks
server UP and checks are
Both the backend and the server may be specified either by their name or
their numeric ID, prefixed with a sharp
This command is restricted and can only be issued on sockets configured
level
get weight
Report the current weight and the initial weight of server <server>
backend <backend> or an error if either doesn't exist. The initial weight
the one that appears in the configuration file. Both are normally
unless the current weight has been changed. Both the backend and the
may be specified either by their name or by their numeric ID, prefixed with
sharp
help
help
Print the list of known keywords and their basic usage. The same help
is also displayed for unknown
prompt
prompt
Toggle the prompt at the beginning of the line and enter or leave
mode. In interactive mode, the connection is not closed after a
completes. Instead, the prompt will appear again, indicating the user
the interpreter is waiting for a new command. The prompt consists in a
angle bracket followed by a space "> ". This mode is particularly
when one wants to periodically check information such as stats or
It is also a good idea to enter interactive mode before issuing a
command.
quit
command.
quit
Close the connection when in interactive
set timeout cli
Change the CLI interface timeout for current connection. This can be
during long debugging sessions where the user needs to constantly
some indicators without being disconnected. The delay is passed in
set weight <backend>/<server>
Change a server's weight to the value passed in argument. If the value
with the '%' sign, then the new weight will be relative to the
configured weight. Relative weights are only permitted between 0 and
and absolute weights are permitted between 0 and 256. Servers which are
of a farm running a static load-balancing algorithm have stricter
because the weight cannot change once set. Thus for these servers, the
accepted values are 0 and 100% (or 0 and the initial weight). Changes
effect immediately, though certain LB algorithms require a certain amount
requests to consider changes. A typical usage of this command is to
a server during an update by setting its weight to zero, then to enable
again after the update by setting it back to 100%. This command is
and can only be issued on sockets configured for level "admin". Both
backend and the server may be specified either by their name or by
numeric ID, prefixed with a sharp
show errors
Dump last known request and response errors collected by frontends
backends. If <iid> is specified, the limit the dump to errors
either frontend or backend whose ID is <iid>. This command is
and can only be issued on sockets configured for levels "operator"
"admin".
"admin".
The errors which may be collected are the last request and response
caused by protocol violations, often due to invalid characters in
names. The report precisely indicates what exact character violated
protocol. Other important information such as the exact date the error
detected, frontend and backend names, the server name (when known),
internal session ID and the source address which has initiated the
are reported
All characters are returned, and non-printable characters are encoded.
most common ones (\t = 9, \n = 10, \r = 13 and \e = 27) are encoded as
letter following a backslash. The backslash itself is encoded as '\\'
avoid confusion. Other non-printable characters are encoded '\xNN'
NN is the two-digits hexadecimal representation of the character's
code.
code.
Lines are prefixed with the position of their first character, starting at
for the beginning of the buffer. At most one input line is printed per
and large lines will be broken into multiple consecutive output lines so
the output never goes beyond 79 characters wide. It is easy to detect if
line was broken, because it will not end with '\n' and the next line's
will be followed by a '+' sign, indicating it is a continuation of
line.
line.
Example
>>> $ echo "show errors" | socat stdio
[04/Mar/2009:15:46:56.081] backend http-in (#2) : invalid
src 127.0.0.1, session #54, frontend fe-eth0 (#1), server s2
response length 213 bytes, error at position
00000 HTTP/1.0 200
00017
00038 Location:
00054 Long-line: this is a very long line which should
00104+ e broken into multiple lines on the output
00154+ otherwise it would be too large to print in a
00204+
00211
In the example above, we see that the backend "http-in" which has
ID 2 has blocked an invalid response from its server s2 which has
ID 1. The request was on session 54 initiated by source 127.0.0.1
received by frontend fe-eth0 whose ID is 1. The total response length
213 bytes when the error was detected, and the error was at byte 23.
is the slash ('/') in header name "header/bizarre", which is not a
HTTP character for a header
show
Dump info about haproxy status on current
show
Dump all known sessions. Avoid doing this on slow connections as this
be huge. This command is restricted and can only be issued on
configured for levels "operator" or
show sess
Display a lot of internal information about the specified session
This identifier is the first field at the beginning of the lines in the
of "show sess" (it corresponds to the session pointer). Those information
useless to most users but may be used by haproxy developers to troubleshoot
complex bug. The output format is intentionally not documented so that it
freely evolve depending on
show stat [<iid> <type>
Dump statistics in the CSV format. By passing <id>, <type> and <sid>, it
possible to dump only selected items
- <iid> is a proxy ID, -1 to dump
- <type> selects the type of dumpable objects : 1 for frontends, 2
backends, 4 for servers, -1 for everything. These values can be
for
1 + 2 = 3 -> frontend +
1 + 2 + 4 = 7 -> frontend + backend +
- <sid> is a server ID, -1 to dump everything from the selected
Example
>>> $ echo "show info;show stat" | socat stdio
Name:
Version: 1.
Release_date:
Nbproc:
Process_num:
(...
# pxname,svname,qcur,qmax,scur,smax,slim,stot,bin,bout,dreq, (...
stats,FRONTEND,,,0,0,1000,0,0,0,0,0,0,,,,,OPEN,,,,,,,,,1,1,0, (...
stats,BACKEND,0,0,0,0,1000,0,0,0,0,0,,0,0,0,0,UP,0,0,0,,0,250,(...
(...
www1,BACKEND,0,0,0,0,1000,0,0,0,0,0,,0,0,0,0,UP,1,1,0,,0,250, (...
$
$
Here, two commands have been issued at once. That way it's easy to
which process the stats apply to in multi-process mode. Notice the
line after the information output which marks the end of the first
A similar empty line appears at the end of the second block (stats) so
the reader knows the output has not been
/*
/*
* Local
* fill-column:
*
*/
*/