This document doesn't provide any configuration help or hints, but it explains
where to find the relevant documents. The summary below is meant to help you
search sections by name and navigate through the document.

Note to documentation contributors :
    This document is formatted with 80 columns per line, with even number of
    spaces for indentation and without tabs. Please follow these rules strictly
    so that it remains easily printable everywhere. If you add sections, please
    update the summary below for easier searching.

The complete HAProxy documentation is contained in the following documents.
Please ensure to consult the relevant documentation to save time and to get the
most accurate response to your needs. Also please refrain from sending questions
to the mailing list whose responses are present in these documents.

  - intro.txt (this document) : it presents the basics of load balancing,
    HAProxy as a product, what it does, what it doesn't do, some known traps to
    avoid, some OS-specific limitations, how to get it, how it evolves, how to
    ensure you're running with all known fixes, how to update it, complements
    and alternatives.

  - management.txt : it explains how to start haproxy, how to manage it at
    runtime, how to manage it on multiple nodes, and how to proceed with
    seamless upgrades.

  - configuration.txt : the reference manual details all configuration keywords
    and their options. It is used when a configuration change is needed.

  - coding-style.txt : this is for developers who want to propose some code to
    the project. It explains the style to adopt for the code. It is not very
    strict and not all the code base completely respects it, but contributions
    which diverge too much from it will be rejected.

  - proxy-protocol.txt : this is the de-facto specification of the PROXY
    protocol which is implemented by HAProxy and a number of third party
    products.

  - README : how to build HAProxy from sources

Load balancing consists in aggregating multiple components in order to achieve
a total processing capacity above each component's individual capacity, without
any intervention from the end user and in a scalable way. This results in more
operations being performed simultaneously by the time it takes a component to
perform only one. A single operation however will still be performed on a single
component at a time and will not get faster than without load balancing. It
always requires at least as many operations as available components and an
efficient load balancing mechanism to make use of all components and to fully
benefit from the load balancing. A good example of this is the number of lanes
on a highway which allows as many cars to pass during the same time frame
without increasing their individual speed.

Examples of load balancing :

  - Process scheduling in multi-processor systems
  - Link load balancing (e.g. EtherChannel, Bonding)
  - IP address load balancing (e.g. ECMP, DNS round-robin)
  - Server load balancing (via load balancers)

The mechanism or component which performs the load balancing operation is
called a load balancer. In web environments these components are called a
"network load balancer", and more commonly a "load balancer" given that this
activity is by far the best known case of load balancing.

A load balancer may act :

  - at the link level : this is called link load balancing, and it consists in
    choosing what network link to send a packet to;

  - at the network level : this is called network load balancing, and it
    consists in choosing what route a series of packets will follow;

  - at the server level : this is called server load balancing and it consists
    in deciding what server will process a connection or request.

Two distinct technologies exist and address different needs, though with some
overlapping. In each case it is important to keep in mind that load balancing
consists in diverting the traffic from its natural flow and that doing so always
requires a minimum of care to maintain the required level of consistency between
all routing decisions.

The first one acts at the packet level and processes packets more or less
individually. There is a 1-to-1 relation between input and output packets, so
it is possible to follow the traffic on both sides of the load balancer using a
regular network sniffer. This technology can be very cheap and extremely fast.
It is usually implemented in hardware (ASICs) allowing to reach line rate, such
as switches doing ECMP. Usually stateless, it can also be stateful (consider
the session a packet belongs to and called layer4-LB or L4), may support DSR
(direct server return, without passing through the LB again) if the packets
were not modified, but provides almost no content awareness. This technology is
very well suited to network-level load balancing, though it is sometimes used
for very basic server load balancing at high speed.

The second one acts on session contents. It requires that the input streams is
reassembled and processed as a whole. The contents may be modified, and the
output stream is segmented into new packets. For this reason it is generally
performed by proxies and they're often called layer 7 load balancers or L7.
This implies that there are two distinct connections on each side, and that
there is no relation between input and output packets sizes nor counts. Clients
and servers are not required to use the same protocol (for example IPv4 vs
IPv6, clear vs SSL). The operations are always stateful, and the return traffic
must pass through the load balancer. The extra processing comes with a cost so
it's not always possible to achieve line rate, especially with small packets.
On the other hand, it offers wide possibilities and is generally achieved by
pure software, even if embedded into hardware appliances. This technology is
very well suited for server load balancing.

Packet-based load balancers are generally deployed in cut-through mode, so they
are installed on the normal path of the traffic and divert it according to the
configuration. The return traffic doesn't necessarily pass through the load
balancer. Some modifications may be applied to the network destination address
in order to direct the traffic to the proper destination. In this case, it is
mandatory that the return traffic passes through the load balancer. If the
routes doesn't make this possible, the load balancer may also replace the
packets' source address with its own in order to force the return traffic to
pass through it.

Proxy-based load balancers are deployed as a server with their own IP addresses
and ports, without architecture changes. Sometimes this requires to perform some
adaptations to the applications so that clients are properly directed to the
load balancer's IP address and not directly to the server's. Some load balancers
may have to adjust some servers' responses to make this possible (e.g. the HTTP
Location header field used in HTTP redirects). Some proxy-based load balancers
may intercept traffic for an address they don't own, and spoof the client's
address when connecting to the server. This allows them to be deployed as if
they were a regular router or firewall, in a cut-through mode very similar to
the packet based load balancers. This is particularly appreciated for products
which combine both packet mode and proxy mode. In this case DSR is obviously
still not possible and the return traffic still has to be routed back to the
load balancer.

A very scalable layered approach would consist in having a front router which
receives traffic from multiple load balanced links, and uses ECMP to distribute
this traffic to a first layer of multiple stateful packet-based load balancers
(L4). These L4 load balancers in turn pass the traffic to an even larger number
of proxy-based load balancers (L7), which have to parse the contents to decide
what server will ultimately receive the traffic.

The number of components and possible paths for the traffic increases the risk
of failure; in very large environments, it is even normal to permanently have
a few faulty components being fixed or replaced. Load balancing done without
awareness of the whole stack's health significantly degrades availability. For
this reason, any sane load balancer will verify that the components it intends
to deliver the traffic to are still alive and reachable, and it will stop
delivering traffic to faulty ones. This can be achieved using various methods.

The most common one consists in periodically sending probes to ensure the
component is still operational. These probes are called "health checks". They
must be representative of the type of failure to address. For example a ping-
based check will not detect that a web server has crashed and doesn't listen to
a port anymore, while a connection to the port will verify this, and a more
advanced request may even validate that the server still works and that the
database it relies on is still accessible. Health checks often involve a few
retries to cover for occasional measuring errors. The period between checks
must be small enough to ensure the faulty component is not used for too long
after an error occurs.

Other methods consist in sampling the production traffic sent to a destination
to observe if it is processed correctly or not, and to evict the components
which return inappropriate responses. However this requires to sacrifice a part
of the production traffic and this is not always acceptable. A combination of
these two mechanisms provides the best of both worlds, with both of them being
used to detect a fault, and only health checks to detect the end of the fault.
A last method involves centralized reporting : a central monitoring agent
periodically updates all load balancers about all components' state. This gives
a global view of the infrastructure to all components, though sometimes with
less accuracy or responsiveness. It's best suited for environments with many
load balancers and many servers.

Layer 7 load balancers also face another challenge known as stickiness or
persistence. The principle is that they generally have to direct multiple
subsequent requests or connections from a same origin (such as an end user) to
the same target. The best known example is the shopping cart on an online
store. If each click leads to a new connection, the user must always be sent
to the server which holds his shopping cart. Content-awareness makes it easier
to spot some elements in the request to identify the server to deliver it to,
but that's not always enough. For example if the source address is used as a
key to pick a server, it can be decided that a hash-based algorithm will be
used and that a given IP address will always be sent to the same server based
on a divide of the address by the number of available servers. But if one
server fails, the result changes and all users are suddenly sent to a different
server and lose their shopping cart. The solution against this issue consists
in memorizing the chosen target so that each time the same visitor is seen,
he's directed to the same server regardless of the number of available servers.
The information may be stored in the load balancer's memory, in which case it
may have to be replicated to other load balancers if it's not alone, or it may
be stored in the client's memory using various methods provided that the client
is able to present this information back with every request (cookie insertion,
redirection to a sub-domain, etc). This mechanism provides the extra benefit of
not having to rely on unstable or unevenly distributed information (such as the
source IP address). This is in fact the strongest reason to adopt a layer 7
load balancer instead of a layer 4 one.

In order to extract information such as a cookie, a host header field, a URL
or whatever, a load balancer may need to decrypt SSL/TLS traffic and even
possibly to re-encrypt it when passing it to the server. This expensive task
explains why in some high-traffic infrastructures, sometimes there may be a
lot of load balancers.

Since a layer 7 load balancer may perform a number of complex operations on the
traffic (decrypt, parse, modify, match cookies, decide what server to send to,
etc), it can definitely cause some trouble and will very commonly be accused of
being responsible for a lot of trouble that it only revealed. Often it will be
discovered that servers are unstable and periodically go up and down, or for
web servers, that they deliver pages with some hard-coded links forcing the
clients to connect directly to one specific server without passing via the load
balancer, or that they take ages to respond under high load causing timeouts.
That's why logging is an extremely important aspect of layer 7 load balancing.
Once a trouble is reported, it is important to figure if the load balancer took
a wrong decision and if so why so that it doesn't happen anymore.

HAProxy is written as "HAProxy" to designate the product, and as "haproxy" to
designate the executable program, software package or a process. However, both
are commonly used for both purposes, and are pronounced H-A-Proxy. Very early,
"haproxy" used to stand for "high availability proxy" and the name was written
in two separate words, though by now it means nothing else than "HAProxy".

HAProxy is :

  - a TCP proxy : it can accept a TCP connection from a listening socket,
    connect to a server and attach these sockets together allowing traffic to
    flow in both directions; IPv4, IPv6 and even UNIX sockets are supported on
    either side, so this can provide an easy way to translate addresses between
    different families.

  - an HTTP reverse-proxy (called a "gateway" in HTTP terminology) : it presents
    itself as a server, receives HTTP requests over connections accepted on a
    listening TCP socket, and passes the requests from these connections to
    servers using different connections. It may use any combination of HTTP/1.x
    or HTTP/2 on any side and will even automatically detect the protocol
    spoken on each side when ALPN is used over TLS.

  - an SSL terminator / initiator / offloader : SSL/TLS may be used on the
    connection coming from the client, on the connection going to the server,
    or even on both connections. A lot of settings can be applied per name
    (SNI), and may be updated at runtime without restarting. Such setups are
    extremely scalable and deployments involving tens to hundreds of thousands
    of certificates were reported.

  - a TCP normalizer : since connections are locally terminated by the operating
    system, there is no relation between both sides, so abnormal traffic such as
    invalid packets, flag combinations, window advertisements, sequence numbers,
    incomplete connections (SYN floods), or so will not be passed to the other
    side. This protects fragile TCP stacks from protocol attacks, and also
    allows to optimize the connection parameters with the client without having
    to modify the servers' TCP stack settings.

  - an HTTP normalizer : when configured to process HTTP traffic, only valid
    complete requests are passed. This protects against a lot of protocol-based
    attacks. Additionally, protocol deviations for which there is a tolerance
    in the specification are fixed so that they don't cause problem on the
    servers (e.g. multiple-line headers).

  - an HTTP fixing tool : it can modify / fix / add / remove / rewrite the URL
    or any request or response header. This helps fixing interoperability issues
    in complex environments.

  - a content-based switch : it can consider any element from the request to
    decide what server to pass the request or connection to. Thus it is possible
    to handle multiple protocols over a same port (e.g. HTTP, HTTPS, SSH).

  - a server load balancer : it can load balance TCP connections and HTTP
    requests. In TCP mode, load balancing decisions are taken for the whole
    connection. In HTTP mode, decisions are taken per request.

  - a traffic regulator : it can apply some rate limiting at various points,
    protect the servers against overloading, adjust traffic priorities based on
    the contents, and even pass such information to lower layers and outer
    network components by marking packets.

  - a protection against DDoS and service abuse : it can maintain a wide number
    of statistics per IP address, URL, cookie, etc and detect when an abuse is
    happening, then take action (slow down the offenders, block them, send them
    to outdated contents, etc).

  - an observation point for network troubleshooting : due to the precision of
    the information reported in logs, it is often used to narrow down some
    network-related issues.

  - an HTTP compression offloader : it can compress responses which were not
    compressed by the server, thus reducing the page load time for clients with
    poor connectivity or using high-latency, mobile networks.

  - a caching proxy : it may cache responses in RAM so that subsequent requests
    for the same object avoid the cost of another network transfer from the
    server as long as the object remains present and valid. It will however not
    store objects to any persistent storage. Please note that this caching
    feature is designed to be maintenance free and focuses solely on saving
    haproxy's precious resources and not on save the server's resources. Caches
    designed to optimize servers require much more tuning and flexibility. If
    you instead need such an advanced cache, please use Varnish Cache, which
    integrates perfectly with haproxy, especially when SSL/TLS is needed on any
    side.

  - a FastCGI gateway : FastCGI can be seen as a different representation of
    HTTP, and as such, HAProxy can directly load-balance a farm comprising any
    combination of FastCGI application servers without requiring to insert
    another level of gateway between them. This results in resource savings and
    a reduction of maintenance costs.

HAProxy is not :

  - an explicit HTTP proxy, i.e. the proxy that browsers use to reach the
    internet. There are excellent open-source software dedicated for this task,
    such as Squid. However HAProxy can be installed in front of such a proxy to
    provide load balancing and high availability.

  - a data scrubber : it will not modify the body of requests nor responses.

  - a static web server : during startup, it isolates itself inside a chroot
    jail and drops its privileges, so that it will not perform any single file-
    system access once started. As such it cannot be turned into a static web
    server (dynamic servers are supported through FastCGI however). There are
    excellent open-source software for this such as Apache or Nginx, and
    HAProxy can be easily installed in front of them to provide load balancing,
    high availability and acceleration.

  - a packet-based load balancer : it will not see IP packets nor UDP datagrams,
    will not perform NAT or even less DSR. These are tasks for lower layers.
    Some kernel-based components such as IPVS (Linux Virtual Server) already do
    this pretty well and complement perfectly with HAProxy.

HAProxy is an event-driven, non-blocking engine combining a very fast I/O layer
with a priority-based, multi-threaded scheduler. As it is designed with a data
forwarding goal in mind, its architecture is optimized to move data as fast as
possible with the least possible operations. It focuses on optimizing the CPU
cache's efficiency by sticking connections to the same CPU as long as possible.
As such it implements a layered model offering bypass mechanisms at each level
ensuring data doesn't reach higher levels unless needed. Most of the processing
is performed in the kernel, and HAProxy does its best to help the kernel do the
work as fast as possible by giving some hints or by avoiding certain operation
when it guesses they could be grouped later. As a result, typical figures show
15% of the processing time spent in HAProxy versus 85% in the kernel in TCP or
HTTP close mode, and about 30% for HAProxy versus 70% for the kernel in HTTP
keep-alive mode.

A single process can run many proxy instances; configurations as large as
300000 distinct proxies in a single process were reported to run fine. A single
core, single CPU setup is far more than enough for more than 99% users, and as
such, users of containers and virtual machines are encouraged to use the
absolute smallest images they can get to save on operational costs and simplify
troubleshooting. However the machine HAProxy runs on must never ever swap, and
its CPU must not be artificially throttled (sub-CPU allocation in hypervisors)
nor be shared with compute-intensive processes which would induce a very high
context-switch latency.

Threading allows to exploit all available processing capacity by using one
thread per CPU core. This is mostly useful for SSL or when data forwarding
rates above 40 Gbps are needed. In such cases it is critically important to
avoid communications between multiple physical CPUs, which can cause strong
bottlenecks in the network stack and in HAProxy itself. While counter-intuitive
to some, the first thing to do when facing some performance issues is often to
reduce the number of CPUs HAProxy runs on.

HAProxy only requires the haproxy executable and a configuration file to run.
For logging it is highly recommended to have a properly configured syslog daemon
and log rotations in place. Logs may also be sent to stdout/stderr, which can be
useful inside containers. The configuration files are parsed before starting,
then HAProxy tries to bind all listening sockets, and refuses to start if
anything fails. Past this point it cannot fail anymore. This means that there
are no runtime failures and that if it accepts to start, it will work until it
is stopped.

Once HAProxy is started, it does exactly 3 things :

  - process incoming connections;

  - periodically check the servers' status (known as health checks);

  - exchange information with other haproxy nodes.

Processing incoming connections is by far the most complex task as it depends
on a lot of configuration possibilities, but it can be summarized as the 9 steps
below :

  - accept incoming connections from listening sockets that belong to a
    configuration entity known as a "frontend", which references one or multiple
    listening addresses;

  - apply the frontend-specific processing rules to these connections that may
    result in blocking them, modifying some headers, or intercepting them to
    execute some internal applets such as the statistics page or the CLI;

  - pass these incoming connections to another configuration entity representing
    a server farm known as a "backend", which contains the list of servers and
    the load balancing strategy for this server farm;

  - apply the backend-specific processing rules to these connections;

  - decide which server to forward the connection to according to the load
    balancing strategy;

  - apply the backend-specific processing rules to the response data;

  - apply the frontend-specific processing rules to the response data;

  - emit a log to report what happened in fine details;

  - in HTTP, loop back to the second step to wait for a new request, otherwise
    close the connection.

Frontends and backends are sometimes considered as half-proxies, since they only
look at one side of an end-to-end connection; the frontend only cares about the
clients while the backend only cares about the servers. HAProxy also supports
full proxies which are exactly the union of a frontend and a backend. When HTTP
processing is desired, the configuration will generally be split into frontends
and backends as they open a lot of possibilities since any frontend may pass a
connection to any backend. With TCP-only proxies, using frontends and backends
rarely provides a benefit and the configuration can be more readable with full
proxies.

This section will enumerate a number of features that HAProxy implements, some
of which are generally expected from any modern load balancer, and some of
which are a direct benefit of HAProxy's architecture. More advanced features
will be detailed in the next section.

Proxying is the action of transferring data between a client and a server over
two independent connections. The following basic features are supported by
HAProxy regarding proxying and connection management :

  - Provide the server with a clean connection to protect them against any
    client-side defect or attack;

  - Listen to multiple IP addresses and/or ports, even port ranges;

  - Transparent accept : intercept traffic targeting any arbitrary IP address
    that doesn't even belong to the local system;

  - Server port doesn't need to be related to listening port, and may even be
    translated by a fixed offset (useful with ranges);

  - Transparent connect : spoof the client's (or any) IP address if needed
    when connecting to the server;

  - Provide a reliable return IP address to the servers in multi-site LBs;

  - Offload the server thanks to buffers and possibly short-lived connections
    to reduce their concurrent connection count and their memory footprint;

  - Optimize TCP stacks (e.g. SACK), congestion control, and reduce RTT impacts;

  - Support different protocol families on both sides (e.g. IPv4/IPv6/Unix);

  - Timeout enforcement : HAProxy supports multiple levels of timeouts depending
    on the stage the connection is, so that a dead client or server, or an
    attacker cannot be granted resources for too long;

  - Protocol validation: HTTP, SSL, or payload are inspected and invalid
    protocol elements are rejected, unless instructed to accept them anyway;

  - Policy enforcement : ensure that only what is allowed may be forwarded;

  - Both incoming and outgoing connections may be limited to certain network
    namespaces (Linux only), making it easy to build a cross-container,
    multi-tenant load balancer;

  - PROXY protocol presents the client's IP address to the server even for
    non-HTTP traffic. This is an HAProxy extension that was adopted by a number
    of third-party products by now, at least these ones at the time of writing :
      - client : haproxy, stud, stunnel, exaproxy, ELB, squid
      - server : haproxy, stud, postfix, exim, nginx, squid, node.js, varnish

HAProxy's SSL stack is recognized as one of the most featureful according to
Google's engineers (http://istlsfastyet.com/). The most commonly used features
making it quite complete are :

  - SNI-based multi-hosting with no limit on sites count and focus on
    performance. At least one deployment is known for running 50000 domains
    with their respective certificates;

  - support for wildcard certificates reduces the need for many certificates ;

  - certificate-based client authentication with configurable policies on
    failure to present a valid certificate. This allows to present a different
    server farm to regenerate the client certificate for example;

  - authentication of the backend server ensures the backend server is the real
    one and not a man in the middle;

  - authentication with the backend server lets the backend server know it's
    really the expected haproxy node that is connecting to it;

  - TLS NPN and ALPN extensions make it possible to reliably offload SPDY/HTTP2
    connections and pass them in clear text to backend servers;

  - OCSP stapling further reduces first page load time by delivering inline an
    OCSP response when the client requests a Certificate Status Request;

  - Dynamic record sizing provides both high performance and low latency, and
    significantly reduces page load time by letting the browser start to fetch
    new objects while packets are still in flight;

  - permanent access to all relevant SSL/TLS layer information for logging,
    access control, reporting etc. These elements can be embedded into HTTP
    header or even as a PROXY protocol extension so that the offloaded server
    gets all the information it would have had if it performed the SSL
    termination itself.

  - Detect, log and block certain known attacks even on vulnerable SSL libs,
    such as the Heartbleed attack affecting certain versions of OpenSSL.

  - support for stateless session resumption (RFC 5077 TLS Ticket extension).
    TLS tickets can be updated from CLI which provides them means to implement
    Perfect Forward Secrecy by frequently rotating the tickets.

HAProxy focuses a lot on availability. As such it cares about servers state,
and about reporting its own state to other network components :

  - Servers' state is continuously monitored using per-server parameters. This
    ensures the path to the server is operational for regular traffic;

  - Health checks support two hysteresis for up and down transitions in order
    to protect against state flapping;

  - Checks can be sent to a different address/port/protocol : this makes it
    easy to check a single service that is considered representative of multiple
    ones, for example the HTTPS port for an HTTP+HTTPS server.

  - Servers can track other servers and go down simultaneously : this ensures
    that servers hosting multiple services can fail atomically and that no one
    will be sent to a partially failed server;

  - Agents may be deployed on the server to monitor load and health : a server
    may be interested in reporting its load, operational status, administrative
    status independently from what health checks can see. By running a simple
    agent on the server, it's possible to consider the server's view of its own
    health in addition to the health checks validating the whole path;

  - Various check methods are available : TCP connect, HTTP request, SMTP hello,
    SSL hello, LDAP, SQL, Redis, send/expect scripts, all with/without SSL;

  - State change is notified in the logs and stats page with the failure reason
    (e.g. the HTTP response received at the moment the failure was detected). An
    e-mail can also be sent to a configurable address upon such a change ;

  - Server state is also reported on the stats interface and can be used to take
    routing decisions so that traffic may be sent to different farms depending
    on their sizes and/or health (e.g. loss of an inter-DC link);

  - HAProxy can use health check requests to pass information to the servers,
    such as their names, weight, the number of other servers in the farm etc.
    so that servers can adjust their response and decisions based on this
    knowledge (e.g. postpone backups to keep more CPU available);

  - Servers can use health checks to report more detailed state than just on/off
    (e.g. I would like to stop, please stop sending new visitors);

  - HAProxy itself can report its state to external components such as routers
    or other load balancers, allowing to build very complete multi-path and
    multi-layer infrastructures.

Just like any serious load balancer, HAProxy cares a lot about availability to
ensure the best global service continuity :

  - Only valid servers are used ; the other ones are automatically evicted from
    load balancing farms ; under certain conditions it is still possible to
    force to use them though;

  - Support for a graceful shutdown so that it is possible to take servers out
    of a farm without affecting any connection;

  - Backup servers are automatically used when active servers are down and
    replace them so that sessions are not lost when possible. This also allows
    to build multiple paths to reach the same server (e.g. multiple interfaces);

  - Ability to return a global failed status for a farm when too many servers
    are down. This, combined with the monitoring capabilities makes it possible
    for an upstream component to choose a different LB node for a given service;

  - Stateless design makes it easy to build clusters : by design, HAProxy does
    its best to ensure the highest service continuity without having to store
    information that could be lost in the event of a failure. This ensures that
    a takeover is the most seamless possible;

  - Integrates well with standard VRRP daemon keepalived : HAProxy easily tells
    keepalived about its state and copes very well with floating virtual IP
    addresses. Note: only use IP redundancy protocols (VRRP/CARP) over cluster-
    based solutions (Heartbeat, ...) as they're the ones offering the fastest,
    most seamless, and most reliable switchover.

HAProxy offers a fairly complete set of load balancing features, most of which
are unfortunately not available in a number of other load balancing products :

  - no less than 10 load balancing algorithms are supported, some of which apply
    to input data to offer an infinite list of possibilities. The most common
    ones are round-robin (for short connections, pick each server in turn),
    leastconn (for long connections, pick the least recently used of the servers
    with the lowest connection count), source (for SSL farms or terminal server
    farms, the server directly depends on the client's source address), URI (for
    HTTP caches, the server directly depends on the HTTP URI), hdr (the server
    directly depends on the contents of a specific HTTP header field), first
    (for short-lived virtual machines, all connections are packed on the
    smallest possible subset of servers so that unused ones can be powered
    down);

  - all algorithms above support per-server weights so that it is possible to
    accommodate from different server generations in a farm, or direct a small
    fraction of the traffic to specific servers (debug mode, running the next
    version of the software, etc);

  - dynamic weights are supported for round-robin, leastconn and consistent
    hashing ; this allows server weights to be modified on the fly from the CLI
    or even by an agent running on the server;

  - slow-start is supported whenever a dynamic weight is supported; this allows
    a server to progressively take the traffic. This is an important feature
    for fragile application servers which require to compile classes at runtime
    as well as cold caches which need to fill up before being run at full
    throttle;

  - hashing can apply to various elements such as client's source address, URL
    components, query string element, header field values, POST parameter, RDP
    cookie;

  - consistent hashing protects server farms against massive redistribution when
    adding or removing servers in a farm. That's very important in large cache
    farms and it allows slow-start to be used to refill cold caches;

  - a number of internal metrics such as the number of connections per server,
    per backend, the amount of available connection slots in a backend etc makes
    it possible to build very advanced load balancing strategies.

Application load balancing would be useless without stickiness. HAProxy provides
a fairly comprehensive set of possibilities to maintain a visitor on the same
server even across various events such as server addition/removal, down/up
cycles, and some methods are designed to be resistant to the distance between
multiple load balancing nodes in that they don't require any replication :

  - stickiness information can be individually matched and learned from
    different places if desired. For example a JSESSIONID cookie may be matched
    both in a cookie and in the URL. Up to 8 parallel sources can be learned at
    the same time and each of them may point to a different stick-table;

  - stickiness information can come from anything that can be seen within a
    request or response, including source address, TCP payload offset and
    length, HTTP query string elements, header field values, cookies, and so
    on.

  - stick-tables are replicated between all nodes in a multi-master fashion;

  - commonly used elements such as SSL-ID or RDP cookies (for TSE farms) are
    directly accessible to ease manipulation;

  - all sticking rules may be dynamically conditioned by ACLs;

  - it is possible to decide not to stick to certain servers, such as backup
    servers, so that when the nominal server comes back, it automatically takes
    the load back. This is often used in multi-path environments;

  - in HTTP it is often preferred not to learn anything and instead manipulate
    a cookie dedicated to stickiness. For this, it's possible to detect,
    rewrite, insert or prefix such a cookie to let the client remember what
    server was assigned;

  - the server may decide to change or clean the stickiness cookie on logout,
    so that leaving visitors are automatically unbound from the server;

  - using ACL-based rules it is also possible to selectively ignore or enforce
    stickiness regardless of the server's state; combined with advanced health
    checks, that helps admins verify that the server they're installing is up
    and running before presenting it to the whole world;

  - an innovative mechanism to set a maximum idle time and duration on cookies
    ensures that stickiness can be smoothly stopped on devices which are never
    closed (smartphones, TVs, home appliances) without having to store them on
    persistent storage;

  - multiple server entries may share the same stickiness keys so that
    stickiness is not lost in multi-path environments when one path goes down;

  - soft-stop ensures that only users with stickiness information will continue
    to reach the server they've been assigned to but no new users will go there.

Logging is an extremely important feature for a load balancer, first because a
load balancer is often wrongly accused of causing the problems it reveals, and
second because it is placed at a critical point in an infrastructure where all
normal and abnormal activity needs to be analyzed and correlated with other
components.

HAProxy provides very detailed logs, with millisecond accuracy and the exact
connection accept time that can be searched in firewalls logs (e.g. for NAT
correlation). By default, TCP and HTTP logs are quite detailed and contain
everything needed for troubleshooting, such as source IP address and port,
frontend, backend, server, timers (request receipt duration, queue duration,
connection setup time, response headers time, data transfer time), global
process state, connection counts, queue status, retries count, detailed
stickiness actions and disconnect reasons, header captures with a safe output
encoding. It is then possible to extend or replace this format to include any
sampled data, variables, captures, resulting in very detailed information. For
example it is possible to log the number of cumulative requests or number of
different URLs visited by a client.

The log level may be adjusted per request using standard ACLs, so it is possible
to automatically silent some logs considered as pollution and instead raise
warnings when some abnormal behavior happen for a small part of the traffic
(e.g. too many URLs or HTTP errors for a source address). Administrative logs
are also emitted with their own levels to inform about the loss or recovery of a
server for example.

Each frontend and backend may use multiple independent log outputs, which eases
multi-tenancy. Logs are preferably sent over UDP, maybe JSON-encoded, and are
truncated after a configurable line length in order to guarantee delivery. But
it is also possible to send them to stdout/stderr or any file descriptor, as
well as to a ring buffer that a client can subscribe to in order to retrieve
them.

HAProxy provides a web-based statistics reporting interface with authentication,
security levels and scopes. It is thus possible to provide each hosted customer
with his own page showing only his own instances. This page can be located in a
hidden URL part of the regular web site so that no new port needs to be opened.
This page may also report the availability of other HAProxy nodes so that it is
easy to spot if everything works as expected at a glance. The view is synthetic
with a lot of details accessible (such as error causes, last access and last
change duration, etc), which are also accessible as a CSV table that other tools
may import to draw graphs. The page may self-refresh to be used as a monitoring
page on a large display. In administration mode, the page also allows to change
server state to ease maintenance operations.

A Prometheus exporter is also provided so that the statistics can be consumed
in a different format depending on the deployment.

In this section, some features that are very commonly used in HAProxy but are
not necessarily present on other load balancers are enumerated.

HAProxy supports information sampling using a wide set of "sample fetch
functions". The principle is to extract pieces of information known as samples,
for immediate use. This is used for stickiness, to build conditions, to produce
information in logs or to enrich HTTP headers.

Samples can be fetched from various sources :

  - constants : integers, strings, IP addresses, binary blocks;

  - the process : date, environment variables, server/frontend/backend/process
    state, byte/connection counts/rates, queue length, random generator, ...

  - variables : per-session, per-request, per-response variables;

  - the client connection : source and destination addresses and ports, and all
    related statistics counters;

  - the SSL client session : protocol, version, algorithm, cipher, key size,
    session ID, all client and server certificate fields, certificate serial,
    SNI, ALPN, NPN, client support for certain extensions;

  - request and response buffers contents : arbitrary payload at offset/length,
    data length, RDP cookie, decoding of SSL hello type, decoding of TLS SNI;

  - HTTP (request and response) : method, URI, path, query string arguments,
    status code, headers values, positional header value, cookies, captures,
    authentication, body elements;

A sample may then pass through a number of operators known as "converters" to
experience some transformation. A converter consumes a sample and produces a
new one, possibly of a completely different type. For example, a converter may
be used to return only the integer length of the input string, or could turn a
string to upper case. Any arbitrary number of converters may be applied in
series to a sample before final use. Among all available sample converters, the
following ones are the most commonly used :

  - arithmetic and logic operators : they make it possible to perform advanced
    computation on input data, such as computing ratios, percentages or simply
    converting from one unit to another one;

  - IP address masks are useful when some addresses need to be grouped by larger
    networks;

  - data representation : URL-decode, base64, hex, JSON strings, hashing;

  - string conversion : extract substrings at fixed positions, fixed length,
    extract specific fields around certain delimiters, extract certain words,
    change case, apply regex-based substitution;

  - date conversion : convert to HTTP date format, convert local to UTC and
    conversely, add or remove offset;

  - lookup an entry in a stick table to find statistics or assigned server;

  - map-based key-to-value conversion from a file (mostly used for geolocation).

Maps are a powerful type of converter consisting in loading a two-columns file
into memory at boot time, then looking up each input sample from the first
column and either returning the corresponding pattern on the second column if
the entry was found, or returning a default value. The output information also
being a sample, it can in turn experience other transformations including other
map lookups. Maps are most commonly used to translate the client's IP address
to an AS number or country code since they support a longest match for network
addresses but they can be used for various other purposes.

Part of their strength comes from being updatable on the fly either from the CLI
or from certain actions using other samples, making them capable of storing and
retrieving information between subsequent accesses. Another strength comes from
the binary tree based indexation which makes them extremely fast even when they
contain hundreds of thousands of entries, making geolocation very cheap and easy
to set up.

Most operations in HAProxy can be made conditional. Conditions are built by
combining multiple ACLs using logic operators (AND, OR, NOT). Each ACL is a
series of tests based on the following elements :

  - a sample fetch method to retrieve the element to test ;

  - an optional series of converters to transform the element ;

  - a list of patterns to match against ;

  - a matching method to indicate how to compare the patterns with the sample

For example, the sample may be taken from the HTTP "Host" header, it could then
be converted to lower case, then matched against a number of regex patterns
using the regex matching method.

Technically, ACLs are built on the same core as the maps, they share the exact
same internal structure, pattern matching methods and performance. The only real
difference is that instead of returning a sample, they only return "found" or
or "not found". In terms of usage, ACL patterns may be declared inline in the
configuration file and do not require their own file. ACLs may be named for ease
of use or to make configurations understandable. A named ACL may be declared
multiple times and it will evaluate all definitions in turn until one matches.

About 13 different pattern matching methods are provided, among which IP address
mask, integer ranges, substrings, regex. They work like functions, and just like
with any programming language, only what is needed is evaluated, so when a
condition involving an OR is already true, next ones are not evaluated, and
similarly when a condition involving an AND is already false, the rest of the
condition is not evaluated.

There is no practical limit to the number of declared ACLs, and a handful of
commonly used ones are provided. However experience has shown that setups using
a lot of named ACLs are quite hard to troubleshoot and that sometimes using
anonymous ACLs inline is easier as it requires less references out of the scope
being analyzed.

HAProxy implements a mechanism known as content-based switching. The principle
is that a connection or request arrives on a frontend, then the information
carried with this request or connection are processed, and at this point it is
possible to write ACLs-based conditions making use of these information to
decide what backend will process the request. Thus the traffic is directed to
one backend or another based on the request's contents. The most common example
consists in using the Host header and/or elements from the path (sub-directories
or file-name extensions) to decide whether an HTTP request targets a static
object or the application, and to route static objects traffic to a backend made
of fast and light servers, and all the remaining traffic to a more complex
application server, thus constituting a fine-grained virtual hosting solution.
This is quite convenient to make multiple technologies coexist as a more global
solution.

Another use case of content-switching consists in using different load balancing
algorithms depending on various criteria. A cache may use a URI hash while an
application would use round-robin.

Last but not least, it allows multiple customers to use a small share of a
common resource by enforcing per-backend (thus per-customer connection limits).

Content switching rules scale very well, though their performance may depend on
the number and complexity of the ACLs in use. But it is also possible to write
dynamic content switching rules where a sample value directly turns into a
backend name and without making use of ACLs at all. Such configurations have
been reported to work fine at least with 300000 backends in production.

Stick-tables are commonly used to store stickiness information, that is, to keep
a reference to the server a certain visitor was directed to. The key is then the
identifier associated with the visitor (its source address, the SSL ID of the
connection, an HTTP or RDP cookie, the customer number extracted from the URL or
from the payload, ...) and the stored value is then the server's identifier.

Stick tables may use 3 different types of samples for their keys : integers,
strings and addresses. Only one stick-table may be referenced in a proxy, and it
is designated everywhere with the proxy name. Up to 8 keys may be tracked in
parallel. The server identifier is committed during request or response
processing once both the key and the server are known.

Stick-table contents may be replicated in active-active mode with other HAProxy
nodes known as "peers" as well as with the new process during a reload operation
so that all load balancing nodes share the same information and take the same
routing decision if client's requests are spread over multiple nodes.

Since stick-tables are indexed on what allows to recognize a client, they are
often also used to store extra information such as per-client statistics. The
extra statistics take some extra space and need to be explicitly declared. The
type of statistics that may be stored includes the input and output bandwidth,
the number of concurrent connections, the connection rate and count over a
period, the amount and frequency of errors, some specific tags and counters,
etc. In order to support keeping such information without being forced to
stick to a given server, a special "tracking" feature is implemented and allows
to track up to 3 simultaneous keys from different tables at the same time
regardless of stickiness rules. Each stored statistics may be searched, dumped
and cleared from the CLI and adds to the live troubleshooting capabilities.

While this mechanism can be used to surclass a returning visitor or to adjust
the delivered quality of service depending on good or bad behavior, it is
mostly used to fight against service abuse and more generally DDoS as it allows
to build complex models to detect certain bad behaviors at a high processing
speed.

There are many places where HAProxy needs to manipulate character strings, such
as logs, redirects, header additions, and so on. In order to provide the
greatest flexibility, the notion of Formatted strings was introduced, initially
for logging purposes, which explains why it's still called "log-format". These
strings contain escape characters allowing to introduce various dynamic data
including variables and sample fetch expressions into strings, and even to
adjust the encoding while the result is being turned into a string (for example,
adding quotes). This provides a powerful way to build header contents, to build
response data or even response templates, or to customize log lines.
Additionally, in order to remain simple to build most common strings, about 50
special tags are provided as shortcuts for information commonly used in logs.

Installing a load balancer in front of an application that was never designed
for this can be a challenging task without the proper tools. One of the most
commonly requested operation in this case is to adjust requests and response
headers to make the load balancer appear as the origin server and to fix hard
coded information. This comes with changing the path in requests (which is
strongly advised against), modifying Host header field, modifying the Location
response header field for redirects, modifying the path and domain attribute
for cookies, and so on. It also happens that a number of servers are somewhat
verbose and tend to leak too much information in the response, making them more
vulnerable to targeted attacks. While it's theoretically not the role of a load
balancer to clean this up, in practice it's located at the best place in the
infrastructure to guarantee that everything is cleaned up.

Similarly, sometimes the load balancer will have to intercept some requests and
respond with a redirect to a new target URL. While some people tend to confuse
redirects and rewriting, these are two completely different concepts, since the
rewriting makes the client and the server see different things (and disagree on
the location of the page being visited) while redirects ask the client to visit
the new URL so that it sees the same location as the server.

In order to do this, HAProxy supports various possibilities for rewriting and
redirects, among which :

  - regex-based URL and header rewriting in requests and responses. Regex are
    the most commonly used tool to modify header values since they're easy to
    manipulate and well understood;

  - headers may also be appended, deleted or replaced based on formatted strings
    so that it is possible to pass information there (e.g. client side TLS
    algorithm and cipher);

  - HTTP redirects can use any 3xx code to a relative, absolute, or completely
    dynamic (formatted string) URI;

  - HTTP redirects also support some extra options such as setting or clearing
    a specific cookie, dropping the query string, appending a slash if missing,
    and so on;

  - a powerful "return" directive allows to customize every part of a response
    like status, headers, body using dynamic contents or even template files.

  - all operations support ACL-based conditions;

HAProxy does a lot to maximize service availability, and for this it takes
large efforts to protect servers against overloading and attacks. The first
and most important point is that only complete and valid requests are forwarded
to the servers. The initial reason is that HAProxy needs to find the protocol
elements it needs to stay synchronized with the byte stream, and the second
reason is that until the request is complete, there is no way to know if some
elements will change its semantics. The direct benefit from this is that servers
are not exposed to invalid or incomplete requests. This is a very effective
protection against slowloris attacks, which have almost no impact on HAProxy.

Another important point is that HAProxy contains buffers to store requests and
responses, and that by only sending a request to a server when it's complete and
by reading the whole response very quickly from the local network, the server
side connection is used for a very short time and this preserves server
resources as much as possible.

A direct extension to this is that HAProxy can artificially limit the number of
concurrent connections or outstanding requests to a server, which guarantees
that the server will never be overloaded even if it continuously runs at 100% of
its capacity during traffic spikes. All excess requests will simply be queued to
be processed when one slot is released. In the end, this huge resource savings
most often ensures so much better server response times that it ends up actually
being faster than by overloading the server. Queued requests may be redispatched
to other servers, or even aborted in queue when the client aborts, which also
protects the servers against the "reload effect", where each click on "reload"
by a visitor on a slow-loading page usually induces a new request and maintains
the server in an overloaded state.

The slow-start mechanism also protects restarting servers against high traffic
levels while they're still finalizing their startup or compiling some classes.

Regarding the protocol-level protection, it is possible to relax the HTTP parser
to accept non standard-compliant but harmless requests or responses and even to
fix them. This allows bogus applications to be accessible while a fix is being
developed. In parallel, offending messages are completely captured with a
detailed report that help developers spot the issue in the application. The most
dangerous protocol violations are properly detected and dealt with and fixed.
For example malformed requests or responses with two Content-length headers are
either fixed if the values are exactly the same, or rejected if they differ,
since it becomes a security problem. Protocol inspection is not limited to HTTP,
it is also available for other protocols like TLS or RDP.

When a protocol violation or attack is detected, there are various options to
respond to the user, such as returning the common "HTTP 400 bad request",
closing the connection with a TCP reset, or faking an error after a long delay
("tarpit") to confuse the attacker. All of these contribute to protecting the
servers by discouraging the offending client from pursuing an attack that
becomes very expensive to maintain.

HAProxy also proposes some more advanced options to protect against accidental
data leaks and session crossing. Not only it can log suspicious server responses
but it will also log and optionally block a response which might affect a given
visitors' confidentiality. One such example is a cacheable cookie appearing in a
cacheable response and which may result in an intermediary cache to deliver it
to another visitor, causing an accidental session sharing.

HAProxy is designed to remain extremely stable and safe to manage in a regular
production environment. It is provided as a single executable file which doesn't
require any installation process. Multiple versions can easily coexist, meaning
that it's possible (and recommended) to upgrade instances progressively by
order of importance instead of migrating all of them at once. Configuration
files are easily versioned. Configuration checking is done off-line so it
doesn't require to restart a service that will possibly fail. During
configuration checks, a number of advanced mistakes may be detected (e.g. a rule
hiding another one, or stickiness that will not work) and detailed warnings and
configuration hints are proposed to fix them. Backwards configuration file
compatibility goes very far away in time, with version 1.5 still fully
supporting configurations for versions 1.1 written 13 years before, and 1.6
only dropping support for almost unused, obsolete keywords that can be done
differently. The configuration and software upgrade mechanism is smooth and non
disruptive in that it allows old and new processes to coexist on the system,
each handling its own connections. System status, build options, and library
compatibility are reported on startup.

Some advanced features allow an application administrator to smoothly stop a
server, detect when there's no activity on it anymore, then take it off-line,
stop it, upgrade it and ensure it doesn't take any traffic while being upgraded,
then test it again through the normal path without opening it to the public, and
all of this without touching HAProxy at all. This ensures that even complicated
production operations may be done during opening hours with all technical
resources available.

The process tries to save resources as much as possible, uses memory pools to
save on allocation time and limit memory fragmentation, releases payload buffers
as soon as their contents are sent, and supports enforcing strong memory limits
above which connections have to wait for a buffer to become available instead of
allocating more memory. This system helps guarantee memory usage in certain
strict environments.

A command line interface (CLI) is available as a UNIX or TCP socket, to perform
a number of operations and to retrieve troubleshooting information. Everything
done on this socket doesn't require a configuration change, so it is mostly used
for temporary changes. Using this interface it is possible to change a server's
address, weight and status, to consult statistics and clear counters, dump and
clear stickiness tables, possibly selectively by key criteria, dump and kill
client-side and server-side connections, dump captured errors with a detailed
analysis of the exact cause and location of the error, dump, add and remove
entries from ACLs and maps, update TLS shared secrets, apply connection limits
and rate limits on the fly to arbitrary frontends (useful in shared hosting
environments), and disable a specific frontend to release a listening port
(useful when daytime operations are forbidden and a fix is needed nonetheless).
Updating certificates and their configuration on the fly is permitted, as well
as enabling and consulting traces of every processing step of the traffic.

For environments where SNMP is mandatory, at least two agents exist, one is
provided with the HAProxy sources and relies on the Net-SNMP Perl module.
Another one is provided with the commercial packages and doesn't require Perl.
Both are roughly equivalent in terms of coverage.

It is often recommended to install 4 utilities on the machine where HAProxy is
deployed :

  - socat (in order to connect to the CLI, though certain forks of netcat can
    also do it to some extents);

  - halog from the latest HAProxy version : this is the log analysis tool, it
    parses native TCP and HTTP logs extremely fast (1 to 2 GB per second) and
    extracts useful information and statistics such as requests per URL, per
    source address, URLs sorted by response time or error rate, termination
    codes etc. It was designed to be deployed on the production servers to
    help troubleshoot live issues so it has to be there ready to be used;

  - tcpdump : this is highly recommended to take the network traces needed to
    troubleshoot an issue that was made visible in the logs. There is a moment
    where application and haproxy's analysis will diverge and the network traces
    are the only way to say who's right and who's wrong. It's also fairly common
    to detect bugs in network stacks and hypervisors thanks to tcpdump;

  - strace : it is tcpdump's companion. It will report what HAProxy really sees
    and will help sort out the issues the operating system is responsible for
    from the ones HAProxy is responsible for. Strace is often requested when a
    bug in HAProxy is suspected;

Depending on the operating system HAProxy is deployed on, certain extra features
may be available or needed. While it is supported on a number of platforms,
HAProxy is primarily developed on Linux, which explains why some features are
only available on this platform.

The transparent bind and connect features, the support for binding connections
to a specific network interface, as well as the ability to bind multiple
processes to the same IP address and ports are only available on Linux and BSD
systems, though only Linux performs a kernel-side load balancing of the incoming
requests between the available processes.

On Linux, there are also a number of extra features and optimizations including
support for network namespaces (also known as "containers") allowing HAProxy to
be a gateway between all containers, the ability to set the MSS, Netfilter marks
and IP TOS field on the client side connection, support for TCP FastOpen on the
listening side, TCP user timeouts to let the kernel quickly kill connections
when it detects the client has disappeared before the configured timeouts, TCP
splicing to let the kernel forward data between the two sides of a connections
thus avoiding multiple memory copies, the ability to enable the "defer-accept"
bind option to only get notified of an incoming connection once data become
available in the kernel buffers, and the ability to send the request with the
ACK confirming a connect (sometimes called "piggy-back") which is enabled with
the "tcp-smart-connect" option. On Linux, HAProxy also takes great care of
manipulating the TCP delayed ACKs to save as many packets as possible on the
network.

Some systems have an unreliable clock which jumps back and forth in the past
and in the future. This used to happen with some NUMA systems where multiple
processors didn't see the exact same time of day, and recently it became more
common in virtualized environments where the virtual clock has no relation with
the real clock, resulting in huge time jumps (sometimes up to 30 seconds have
been observed). This causes a lot of trouble with respect to timeout enforcement
in general. Due to this flaw of these systems, HAProxy maintains its own
monotonic clock which is based on the system's clock but where drift is measured
and compensated for. This ensures that even with a very bad system clock, timers
remain reasonably accurate and timeouts continue to work. Note that this problem
affects all the software running on such systems and is not specific to HAProxy.
The common effects are spurious timeouts or application freezes. Thus if this
behavior is detected on a system, it must be fixed, regardless of the fact that
HAProxy protects itself against it.

On Linux, a new starting process may communicate with the previous one to reuse
its listening file descriptors so that the listening sockets are never
interrupted during the process's replacement.

HAProxy can be built with support for the Lua embedded language, which opens a
wide area of new possibilities related to complex manipulation of requests or
responses, routing decisions, statistics processing and so on. Using Lua it is
even possible to establish parallel connections to other servers to exchange
information. This way it becomes possible (though complex) to develop an
authentication system for example. Please refer to the documentation in the file
"doc/lua-api/index.rst" for more information on how to use Lua.

At any moment an administrator may connect over the CLI and enable tracing in
various internal subsystems. Various levels of details are provided by default
so that in practice anything between one line per request to 500 lines per
request can be retrieved. Filters as well as an automatic capture on/off/pause
mechanism are available so that it really is possible to wait for a certain
event and watch it in detail. This is extremely convenient to diagnose protocol
violations from faulty servers and clients, or denial of service attacks.

Typical CPU usage figures show 15% of the processing time spent in HAProxy
versus 85% in the kernel in TCP or HTTP close mode, and about 30% for HAProxy
versus 70% for the kernel in HTTP keep-alive mode. This means that the operating
system and its tuning have a strong impact on the global performance.

Usages vary a lot between users, some focus on bandwidth, other ones on request
rate, others on connection concurrency, others on SSL performance. This section
aims at providing a few elements to help with this task.

It is important to keep in mind that every operation comes with a cost, so each
individual operation adds its overhead on top of the other ones, which may be
negligible in certain circumstances, and which may dominate in other cases.

When processing the requests from a connection, we can say that :

  - forwarding data costs less than parsing request or response headers;

  - parsing request or response headers cost less than establishing then closing
    a connection to a server;

  - establishing an closing a connection costs less than a TLS resume operation;

  - a TLS resume operation costs less than a full TLS handshake with a key
    computation;

  - an idle connection costs less CPU than a connection whose buffers hold data;

  - a TLS context costs even more memory than a connection with data;

So in practice, it is cheaper to process payload bytes than header bytes, thus
it is easier to achieve high network bandwidth with large objects (few requests
per volume unit) than with small objects (many requests per volume unit). This
explains why maximum bandwidth is always measured with large objects, while
request rate or connection rates are measured with small objects.

Some operations scale well on multiple processes spread over multiple CPUs,
and others don't scale as well. Network bandwidth doesn't scale very far because
the CPU is rarely the bottleneck for large objects, it's mostly the network
bandwidth and data buses to reach the network interfaces. The connection rate
doesn't scale well over multiple processors due to a few locks in the system
when dealing with the local ports table. The request rate over persistent
connections scales very well as it doesn't involve much memory nor network
bandwidth and doesn't require to access locked structures. TLS key computation
scales very well as it's totally CPU-bound. TLS resume scales moderately well,
but reaches its limits around 4 processes where the overhead of accessing the
shared table offsets the small gains expected from more power.

The performance numbers one can expect from a very well tuned system are in the
following range. It is important to take them as orders of magnitude and to
expect significant variations in any direction based on the processor, IRQ
setting, memory type, network interface type, operating system tuning and so on.

The following numbers were found on a Core i7 running at 3.7 GHz equipped with
a dual-port 10 Gbps NICs running Linux kernel 3.10, HAProxy 1.6 and OpenSSL
1.0.2. HAProxy was running as a single process on a single dedicated CPU core,
and two extra cores were dedicated to network interrupts :

  - 20 Gbps of maximum network bandwidth in clear text for objects 256 kB or
    higher, 10 Gbps for 41kB or higher;

  - 4.6 Gbps of TLS traffic using AES256-GCM cipher with large objects;

  - 83000 TCP connections per second from client to server;

  - 82000 HTTP connections per second from client to server;

  - 97000 HTTP requests per second in server-close mode (keep-alive with the
    client, close with the server);

  - 243000 HTTP requests per second in end-to-end keep-alive mode;

  - 300000 filtered TCP connections per second (anti-DDoS)

  - 160000 HTTPS requests per second in keep-alive mode over persistent TLS
    connections;

  - 13100 HTTPS requests per second using TLS resumed connections;

  - 1300 HTTPS connections per second using TLS connections renegotiated with
    RSA2048;

  - 20000 concurrent saturated connections per GB of RAM, including the memory
    required for system buffers; it is possible to do better with careful tuning
    but this result it easy to achieve.

  - about 8000 concurrent TLS connections (client-side only) per GB of RAM,
    including the memory required for system buffers;

  - about 5000 concurrent end-to-end TLS connections (both sides) per GB of
    RAM including the memory required for system buffers;

A more recent benchmark featuring the multi-thread enabled HAProxy 2.4 on a
64-core ARM Graviton2 processor in AWS reached 2 million HTTPS requests per
second at sub-millisecond response time, and 100 Gbps of traffic:

  https://www.haproxy.com/blog/haproxy-forwards-over-2-million-http-requests-per-second-on-a-single-aws-arm-instance/

Thus a good rule of thumb to keep in mind is that the request rate is divided
by 10 between TLS keep-alive and TLS resume, and between TLS resume and TLS
renegotiation, while it's only divided by 3 between HTTP keep-alive and HTTP
close. Another good rule of thumb is to remember that a high frequency core
with AES instructions can do around 20 Gbps of AES-GCM per core.

Another good rule of thumb is to consider that on the same server, HAProxy will
be able to saturate :

  - about 5-10 static file servers or caching proxies;

  - about 100 anti-virus proxies;

  - and about 100-1000 application servers depending on the technology in use.

HAProxy is an open source project covered by the GPLv2 license, meaning that
everyone is allowed to redistribute it provided that access to the sources is
also provided upon request, especially if any modifications were made.

HAProxy evolves as a main development branch called "master" or "mainline", from
which new branches are derived once the code is considered stable. A lot of web
sites run some development branches in production on a voluntarily basis, either
to participate to the project or because they need a bleeding edge feature, and
their feedback is highly valuable to fix bugs and judge the overall quality and
stability of the version being developed.

The new branches that are created when the code is stable enough constitute a
stable version and are generally maintained for several years, so that there is
no emergency to migrate to a newer branch even when you're not on the latest.
Once a stable branch is issued, it may only receive bug fixes, and very rarely
minor feature updates when that makes users' life easier. All fixes that go into
a stable branch necessarily come from the master branch. This guarantees that no
fix will be lost after an upgrade. For this reason, if you fix a bug, please
make the patch against the master branch, not the stable branch. You may even
discover it was already fixed. This process also ensures that regressions in a
stable branch are extremely rare, so there is never any excuse for not upgrading
to the latest version in your current branch.

Branches are numbered with two digits delimited with a dot, such as "1.6".
Since 1.9, branches with an odd second digit are mostly focused on sensitive
technical updates and more aimed at advanced users because they are likely to
trigger more bugs than the other ones. They are maintained for about a year
only and must not be deployed where they cannot be rolled back in emergency. A
complete version includes one or two sub-version numbers indicating the level of
fix. For example, version 1.5.14 is the 14th fix release in branch 1.5 after
version 1.5.0 was issued. It contains 126 fixes for individual bugs, 24 updates
on the documentation, and 75 other backported patches, most of which were needed
to fix the aforementioned 126 bugs. An existing feature may never be modified
nor removed in a stable branch, in order to guarantee that upgrades within the
same branch will always be harmless.

HAProxy is available from multiple sources, at different release rhythms :

  - The official community web site : http://www.haproxy.org/ : this site
    provides the sources of the latest development release, all stable releases,
    as well as nightly snapshots for each branch. The release cycle is not fast,
    several months between stable releases, or between development snapshots.
    Very old versions are still supported there. Everything is provided as
    sources only, so whatever comes from there needs to be rebuilt and/or
    repackaged;

  - GitHub : https://github.com/haproxy/haproxy/ : this is the mirror for the
    development branch only, which provides integration with the issue tracker,
    continuous integration and code coverage tools. This is exclusively for
    contributors;

  - A number of operating systems such as Linux distributions and BSD ports.
    These systems generally provide long-term maintained versions which do not
    always contain all the fixes from the official ones, but which at least
    contain the critical fixes. It often is a good option for most users who do
    not seek advanced configurations and just want to keep updates easy;

  - Commercial versions from http://www.haproxy.com/ : these are supported
    professional packages built for various operating systems or provided as
    appliances, based on the latest stable versions and including a number of
    features backported from the next release for which there is a strong
    demand. It is the best option for users seeking the latest features with
    the reliability of a stable branch, the fastest response time to fix bugs,
    or simply support contracts on top of an open source product;


In order to ensure that the version you're using is the latest one in your
branch, you need to proceed this way :

  - verify which HAProxy executable you're running : some systems ship it by
    default and administrators install their versions somewhere else on the
    system, so it is important to verify in the startup scripts which one is
    used;

  - determine which source your HAProxy version comes from. For this, it's
    generally sufficient to type "haproxy -v". A development version will
    appear like this, with the "dev" word after the branch number :

      HAProxy version 2.4-dev18-a5357c-137 2021/05/09 - https://haproxy.org/

    A stable version will appear like this, as well as unmodified stable
    versions provided by operating system vendors :

      HAProxy version 1.5.14 2015/07/02

    And a nightly snapshot of a stable version will appear like this with an
    hexadecimal sequence after the version, and with the date of the snapshot
    instead of the date of the release :

      HAProxy version 1.5.14-e4766ba 2015/07/29

    Any other format may indicate a system-specific package with its own
    patch set. For example HAProxy Enterprise versions will appear with the
    following format (<branch>-<latest commit>-<revision>) :

      HAProxy version 1.5.0-994126-357 2015/07/02

    Please note that historically versions prior to 2.4 used to report the
    process name with a hyphen between "HA" and "Proxy", including those above
    which were adjusted to show the correct format only, so better ignore this
    word or use a relaxed match in scripts. Additionally, modern versions add
    a URL linking to the project's home.

    Finally, versions 2.1 and above will include a "Status" line indicating
    whether the version is safe for production or not, and if so, till when, as
    well as a link to the list of known bugs affecting this version.

  - for system-specific packages, you have to check with your vendor's package
    repository or update system to ensure that your system is still supported,
    and that fixes are still provided for your branch. For community versions
    coming from haproxy.org, just visit the site, verify the status of your
    branch and compare the latest version with yours to see if you're on the
    latest one. If not you can upgrade. If your branch is not maintained
    anymore, you're definitely very late and will have to consider an upgrade
    to a more recent branch (carefully read the README when doing so).

HAProxy will have to be updated according to the source it came from. Usually it
follows the system vendor's way of upgrading a package. If it was taken from
sources, please read the README file in the sources directory after extracting
the sources and follow the instructions for your operating system.

HAProxy integrates fairly well with certain products listed below, which is why
they are mentioned here even if not directly related to HAProxy.

Apache is the de-facto standard HTTP server. It's a very complete and modular
project supporting both file serving and dynamic contents. It can serve as a
frontend for some application servers. It can even proxy requests and cache
responses. In all of these use cases, a front load balancer is commonly needed.
Apache can work in various modes, some being heavier than others. Certain
modules still require the heavier pre-forked model and will prevent Apache from
scaling well with a high number of connections. In this case HAProxy can provide
a tremendous help by enforcing the per-server connection limits to a safe value
and will significantly speed up the server and preserve its resources that will
be better used by the application.

Apache can extract the client's address from the X-Forwarded-For header by using
the "mod_rpaf" extension. HAProxy will automatically feed this header when
"option forwardfor" is specified in its configuration. HAProxy may also offer a
nice protection to Apache when exposed to the internet, where it will better
resist a wide number of types of DoS attacks.

NGINX is the second de-facto standard HTTP server. Just like Apache, it covers a
wide range of features. NGINX is built on a similar model as HAProxy so it has
no problem dealing with tens of thousands of concurrent connections. When used
as a gateway to some applications (e.g. using the included PHP FPM) it can often
be beneficial to set up some frontend connection limiting to reduce the load
on the PHP application. HAProxy will clearly be useful there both as a regular
load balancer and as the traffic regulator to speed up PHP by decongesting
it. Also since both products use very little CPU thanks to their event-driven
architecture, it's often easy to install both of them on the same system. NGINX
implements HAProxy's PROXY protocol, thus it is easy for HAProxy to pass the
client's connection information to NGINX so that the application gets all the
relevant information. Some benchmarks have also shown that for large static
file serving, implementing consistent hash on HAProxy in front of NGINX can be
beneficial by optimizing the OS' cache hit ratio, which is basically multiplied
by the number of server nodes.

Varnish is a smart caching reverse-proxy, probably best described as a web
application accelerator. Varnish doesn't implement SSL/TLS and wants to dedicate
all of its CPU cycles to what it does best. Varnish also implements HAProxy's
PROXY protocol so that HAProxy can very easily be deployed in front of Varnish
as an SSL offloader as well as a load balancer and pass it all relevant client
information. Also, Varnish naturally supports decompression from the cache when
a server has provided a compressed object, but doesn't compress however. HAProxy
can then be used to compress outgoing data when backend servers do not implement
compression, though it's rarely a good idea to compress on the load balancer
unless the traffic is low.

When building large caching farms across multiple nodes, HAProxy can make use of
consistent URL hashing to intelligently distribute the load to the caching nodes
and avoid cache duplication, resulting in a total cache size which is the sum of
all caching nodes. In addition, caching of very small dumb objects for a short
duration on HAProxy can sometimes save network round trips and reduce the CPU
load on both the HAProxy and the Varnish nodes. This is only possible is no
processing is done on these objects on Varnish (this is often referred to as
the notion of "favicon cache", by which a sizeable percentage of useless
downstream requests can sometimes be avoided). However do not enable HAProxy
caching for a long time (more than a few seconds) in front of any other cache,
that would significantly complicate troubleshooting without providing really
significant savings.

Linux Virtual Server (LVS or IPVS) is the layer 4 load balancer included within
the Linux kernel. It works at the packet level and handles TCP and UDP. In most
cases it's more a complement than an alternative since it doesn't have layer 7
knowledge at all.

Pound is another well-known load balancer. It's much simpler and has much less
features than HAProxy but for many very basic setups both can be used. Its
author has always focused on code auditability first and wants to maintain the
set of features low. Its thread-based architecture scales less well with high
connection counts, but it's a good product.

Pen is a quite light load balancer. It supports SSL, maintains persistence using
a fixed-size table of its clients' IP addresses. It supports a packet-oriented
mode allowing it to support direct server return and UDP to some extents. It is
meant for small loads (the persistence table only has 2048 entries).

NGINX can do some load balancing to some extents, though it's clearly not its
primary function. Production traffic is used to detect server failures, the
load balancing algorithms are more limited, and the stickiness is very limited.
But it can make sense in some simple deployment scenarios where it is already
present. The good thing is that since it integrates very well with HAProxy,
there's nothing wrong with adding HAProxy later when its limits have been
reached.

Varnish also does some load balancing of its backend servers and does support
real health checks. It doesn't implement stickiness however, so just like with
NGINX, as long as stickiness is not needed that can be enough to start with.
And similarly, since HAProxy and Varnish integrate so well together, it's easy
to add it later into the mix to complement the feature set.

If you want to contact the developers or any community member about anything,
the best way to do it usually is via the mailing list by sending your message
to haproxy@formilux.org. Please note that this list is public and its archives
are public as well so you should avoid disclosing sensitive information. A
thousand of users of various experience levels are present there and even the
most complex questions usually find an optimal response relatively quickly.
Suggestions are welcome too. For users having difficulties with e-mail, a
Discourse platform is available at http://discourse.haproxy.org/ . However
please keep in mind that there are less people reading questions there and that
most are handled by a really tiny team. In any case, please be patient and
respectful with those who devote their spare time helping others.

I you believe you've found a bug but are not sure, it's best reported on the
mailing list. If you're quite convinced you've found a bug, that your version
is up-to-date in its branch, and you already have a GitHub account, feel free
to go directly to https://github.com/haproxy/haproxy/ and file an issue with
all possibly available details. Again, this is public so be careful not to post
information you might later regret. Since the issue tracker presents itself as
a very long thread, please avoid pasting very long dumps (a few hundreds lines
or more) and attach them instead.

If you've found what you're absolutely certain can be considered a critical
security issue that would put many users in serious trouble if discussed in a
public place, then you can send it with the reproducer to security@haproxy.org.
A small team of trusted developers will receive it and will be able to propose
a fix. We usually don't use embargoes and once a fix is available it gets
merged. In some rare circumstances it can happen that a release is coordinated
with software vendors. Please note that this process usually messes up with
eveyone's work, and that rushed up releases can sometimes introduce new bugs,
so it's best avoided unless strictly necessary; as such, there is often little
consideration for reports that needlessly cause such extra burden, and the best
way to see your work credited usually is to provide a working fix, which will
appear in changelogs.