blob: 3a42ed565b8196a544ee3b8c9a18eadf48fd7493 [file] [log] [blame]
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001 -----------------------
2 HAProxy Starter Guide
3 -----------------------
Willy Tarreau33205c22020-07-07 16:35:28 +02004 version 2.3
Willy Tarreaud8e42b62015-08-18 21:51:36 +02005
6
7This document is an introduction to HAProxy for all those who don't know it, as
8well as for those who want to re-discover it when they know older versions. Its
9primary focus is to provide users with all the elements to decide if HAProxy is
10the product they're looking for or not. Advanced users may find here some parts
11of solutions to some ideas they had just because they were not aware of a given
Davor Ocelic4094ce12017-12-19 23:30:39 +010012new feature. Some sizing information is also provided, the product's lifecycle
Willy Tarreaud8e42b62015-08-18 21:51:36 +020013is explained, and comparisons with partially overlapping products are provided.
14
Davor Ocelic4094ce12017-12-19 23:30:39 +010015This document doesn't provide any configuration help or hints, but it explains
Willy Tarreaud8e42b62015-08-18 21:51:36 +020016where to find the relevant documents. The summary below is meant to help you
17search sections by name and navigate through the document.
18
19Note to documentation contributors :
20 This document is formatted with 80 columns per line, with even number of
21 spaces for indentation and without tabs. Please follow these rules strictly
22 so that it remains easily printable everywhere. If you add sections, please
23 update the summary below for easier searching.
24
25
26Summary
27-------
28
291. Available documentation
30
312. Quick introduction to load balancing and load balancers
32
333. Introduction to HAProxy
343.1. What HAProxy is and is not
353.2. How HAProxy works
363.3. Basic features
373.3.1. Proxying
383.3.2. SSL
393.3.3. Monitoring
403.3.4. High availability
413.3.5. Load balancing
423.3.6. Stickiness
433.3.7. Sampling and converting information
443.3.8. Maps
453.3.9. ACLs and conditions
463.3.10. Content switching
473.3.11. Stick-tables
Davor Ocelic4094ce12017-12-19 23:30:39 +0100483.3.12. Formatted strings
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200493.3.13. HTTP rewriting and redirection
503.3.14. Server protection
513.3.15. Logging
523.3.16. Statistics
533.4. Advanced features
543.4.1. Management
553.4.2. System-specific capabilities
563.4.3. Scripting
573.5. Sizing
583.6. How to get HAProxy
59
604. Companion products and alternatives
614.1. Apache HTTP server
624.2. NGINX
634.3. Varnish
644.4. Alternatives
65
Willy Tarreau65626232020-05-05 18:08:07 +0200665. Contacts
67
Willy Tarreaud8e42b62015-08-18 21:51:36 +020068
691. Available documentation
70--------------------------
71
72The complete HAProxy documentation is contained in the following documents.
73Please ensure to consult the relevant documentation to save time and to get the
74most accurate response to your needs. Also please refrain from sending questions
75to the mailing list whose responses are present in these documents.
76
77 - intro.txt (this document) : it presents the basics of load balancing,
78 HAProxy as a product, what it does, what it doesn't do, some known traps to
79 avoid, some OS-specific limitations, how to get it, how it evolves, how to
Davor Ocelic4094ce12017-12-19 23:30:39 +010080 ensure you're running with all known fixes, how to update it, complements
81 and alternatives.
Willy Tarreaud8e42b62015-08-18 21:51:36 +020082
Willy Tarreau373933d2015-10-13 16:32:20 +020083 - management.txt : it explains how to start haproxy, how to manage it at
Davor Ocelic4094ce12017-12-19 23:30:39 +010084 runtime, how to manage it on multiple nodes, and how to proceed with
85 seamless upgrades.
Willy Tarreau373933d2015-10-13 16:32:20 +020086
Willy Tarreaud8e42b62015-08-18 21:51:36 +020087 - configuration.txt : the reference manual details all configuration keywords
88 and their options. It is used when a configuration change is needed.
89
Willy Tarreaud8e42b62015-08-18 21:51:36 +020090 - coding-style.txt : this is for developers who want to propose some code to
Davor Ocelic4094ce12017-12-19 23:30:39 +010091 the project. It explains the style to adopt for the code. It is not very
92 strict and not all the code base completely respects it, but contributions
Willy Tarreaud8e42b62015-08-18 21:51:36 +020093 which diverge too much from it will be rejected.
94
95 - proxy-protocol.txt : this is the de-facto specification of the PROXY
96 protocol which is implemented by HAProxy and a number of third party
97 products.
98
Davor Ocelic4094ce12017-12-19 23:30:39 +010099 - README : how to build HAProxy from sources
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200100
101
1022. Quick introduction to load balancing and load balancers
103----------------------------------------------------------
104
105Load balancing consists in aggregating multiple components in order to achieve
106a total processing capacity above each component's individual capacity, without
107any intervention from the end user and in a scalable way. This results in more
Willy Tarreaueff04f42015-08-27 14:44:43 +0200108operations being performed simultaneously by the time it takes a component to
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200109perform only one. A single operation however will still be performed on a single
110component at a time and will not get faster than without load balancing. It
111always requires at least as many operations as available components and an
112efficient load balancing mechanism to make use of all components and to fully
113benefit from the load balancing. A good example of this is the number of lanes
114on a highway which allows as many cars to pass during the same time frame
115without increasing their individual speed.
116
117Examples of load balancing :
118
119 - Process scheduling in multi-processor systems
Davor Ocelic4094ce12017-12-19 23:30:39 +0100120 - Link load balancing (e.g. EtherChannel, Bonding)
121 - IP address load balancing (e.g. ECMP, DNS round-robin)
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200122 - Server load balancing (via load balancers)
123
124The mechanism or component which performs the load balancing operation is
125called a load balancer. In web environments these components are called a
126"network load balancer", and more commonly a "load balancer" given that this
127activity is by far the best known case of load balancing.
128
129A load balancer may act :
130
131 - at the link level : this is called link load balancing, and it consists in
Patrick Starrdce734e2017-10-09 13:17:12 +0700132 choosing what network link to send a packet to;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200133
134 - at the network level : this is called network load balancing, and it
Patrick Starrdce734e2017-10-09 13:17:12 +0700135 consists in choosing what route a series of packets will follow;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200136
137 - at the server level : this is called server load balancing and it consists
138 in deciding what server will process a connection or request.
139
140Two distinct technologies exist and address different needs, though with some
Willy Tarreaueff04f42015-08-27 14:44:43 +0200141overlapping. In each case it is important to keep in mind that load balancing
142consists in diverting the traffic from its natural flow and that doing so always
143requires a minimum of care to maintain the required level of consistency between
144all routing decisions.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200145
146The first one acts at the packet level and processes packets more or less
147individually. There is a 1-to-1 relation between input and output packets, so
148it is possible to follow the traffic on both sides of the load balancer using a
Davor Ocelic4094ce12017-12-19 23:30:39 +0100149regular network sniffer. This technology can be very cheap and extremely fast.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200150It is usually implemented in hardware (ASICs) allowing to reach line rate, such
Davor Ocelic4094ce12017-12-19 23:30:39 +0100151as switches doing ECMP. Usually stateless, it can also be stateful (consider
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200152the session a packet belongs to and called layer4-LB or L4), may support DSR
153(direct server return, without passing through the LB again) if the packets
154were not modified, but provides almost no content awareness. This technology is
155very well suited to network-level load balancing, though it is sometimes used
156for very basic server load balancing at high speed.
157
158The second one acts on session contents. It requires that the input streams is
159reassembled and processed as a whole. The contents may be modified, and the
160output stream is segmented into new packets. For this reason it is generally
161performed by proxies and they're often called layer 7 load balancers or L7.
162This implies that there are two distinct connections on each side, and that
163there is no relation between input and output packets sizes nor counts. Clients
164and servers are not required to use the same protocol (for example IPv4 vs
165IPv6, clear vs SSL). The operations are always stateful, and the return traffic
166must pass through the load balancer. The extra processing comes with a cost so
167it's not always possible to achieve line rate, especially with small packets.
168On the other hand, it offers wide possibilities and is generally achieved by
169pure software, even if embedded into hardware appliances. This technology is
170very well suited for server load balancing.
171
172Packet-based load balancers are generally deployed in cut-through mode, so they
173are installed on the normal path of the traffic and divert it according to the
174configuration. The return traffic doesn't necessarily pass through the load
175balancer. Some modifications may be applied to the network destination address
176in order to direct the traffic to the proper destination. In this case, it is
177mandatory that the return traffic passes through the load balancer. If the
178routes doesn't make this possible, the load balancer may also replace the
179packets' source address with its own in order to force the return traffic to
180pass through it.
181
Davor Ocelic4094ce12017-12-19 23:30:39 +0100182Proxy-based load balancers are deployed as a server with their own IP addresses
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200183and ports, without architecture changes. Sometimes this requires to perform some
184adaptations to the applications so that clients are properly directed to the
185load balancer's IP address and not directly to the server's. Some load balancers
Davor Ocelic4094ce12017-12-19 23:30:39 +0100186may have to adjust some servers' responses to make this possible (e.g. the HTTP
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200187Location header field used in HTTP redirects). Some proxy-based load balancers
188may intercept traffic for an address they don't own, and spoof the client's
189address when connecting to the server. This allows them to be deployed as if
190they were a regular router or firewall, in a cut-through mode very similar to
191the packet based load balancers. This is particularly appreciated for products
192which combine both packet mode and proxy mode. In this case DSR is obviously
193still not possible and the return traffic still has to be routed back to the
194load balancer.
195
196A very scalable layered approach would consist in having a front router which
197receives traffic from multiple load balanced links, and uses ECMP to distribute
198this traffic to a first layer of multiple stateful packet-based load balancers
199(L4). These L4 load balancers in turn pass the traffic to an even larger number
200of proxy-based load balancers (L7), which have to parse the contents to decide
201what server will ultimately receive the traffic.
202
203The number of components and possible paths for the traffic increases the risk
204of failure; in very large environments, it is even normal to permanently have
205a few faulty components being fixed or replaced. Load balancing done without
206awareness of the whole stack's health significantly degrades availability. For
207this reason, any sane load balancer will verify that the components it intends
208to deliver the traffic to are still alive and reachable, and it will stop
209delivering traffic to faulty ones. This can be achieved using various methods.
210
211The most common one consists in periodically sending probes to ensure the
212component is still operational. These probes are called "health checks". They
213must be representative of the type of failure to address. For example a ping-
214based check will not detect that a web server has crashed and doesn't listen to
215a port anymore, while a connection to the port will verify this, and a more
216advanced request may even validate that the server still works and that the
217database it relies on is still accessible. Health checks often involve a few
218retries to cover for occasional measuring errors. The period between checks
219must be small enough to ensure the faulty component is not used for too long
220after an error occurs.
221
222Other methods consist in sampling the production traffic sent to a destination
Davor Ocelic4094ce12017-12-19 23:30:39 +0100223to observe if it is processed correctly or not, and to evict the components
Patrick Starrdce734e2017-10-09 13:17:12 +0700224which return inappropriate responses. However this requires to sacrifice a part
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200225of the production traffic and this is not always acceptable. A combination of
226these two mechanisms provides the best of both worlds, with both of them being
227used to detect a fault, and only health checks to detect the end of the fault.
228A last method involves centralized reporting : a central monitoring agent
229periodically updates all load balancers about all components' state. This gives
230a global view of the infrastructure to all components, though sometimes with
231less accuracy or responsiveness. It's best suited for environments with many
232load balancers and many servers.
233
234Layer 7 load balancers also face another challenge known as stickiness or
235persistence. The principle is that they generally have to direct multiple
236subsequent requests or connections from a same origin (such as an end user) to
237the same target. The best known example is the shopping cart on an online
238store. If each click leads to a new connection, the user must always be sent
239to the server which holds his shopping cart. Content-awareness makes it easier
240to spot some elements in the request to identify the server to deliver it to,
241but that's not always enough. For example if the source address is used as a
242key to pick a server, it can be decided that a hash-based algorithm will be
243used and that a given IP address will always be sent to the same server based
244on a divide of the address by the number of available servers. But if one
245server fails, the result changes and all users are suddenly sent to a different
246server and lose their shopping cart. The solution against this issue consists
247in memorizing the chosen target so that each time the same visitor is seen,
248he's directed to the same server regardless of the number of available servers.
249The information may be stored in the load balancer's memory, in which case it
250may have to be replicated to other load balancers if it's not alone, or it may
251be stored in the client's memory using various methods provided that the client
252is able to present this information back with every request (cookie insertion,
253redirection to a sub-domain, etc). This mechanism provides the extra benefit of
254not having to rely on unstable or unevenly distributed information (such as the
255source IP address). This is in fact the strongest reason to adopt a layer 7
256load balancer instead of a layer 4 one.
257
258In order to extract information such as a cookie, a host header field, a URL
259or whatever, a load balancer may need to decrypt SSL/TLS traffic and even
Davor Ocelic4094ce12017-12-19 23:30:39 +0100260possibly to re-encrypt it when passing it to the server. This expensive task
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200261explains why in some high-traffic infrastructures, sometimes there may be a
262lot of load balancers.
263
264Since a layer 7 load balancer may perform a number of complex operations on the
265traffic (decrypt, parse, modify, match cookies, decide what server to send to,
266etc), it can definitely cause some trouble and will very commonly be accused of
267being responsible for a lot of trouble that it only revealed. Often it will be
268discovered that servers are unstable and periodically go up and down, or for
269web servers, that they deliver pages with some hard-coded links forcing the
270clients to connect directly to one specific server without passing via the load
271balancer, or that they take ages to respond under high load causing timeouts.
272That's why logging is an extremely important aspect of layer 7 load balancing.
273Once a trouble is reported, it is important to figure if the load balancer took
274a wrong decision and if so why so that it doesn't happen anymore.
275
276
2773. Introduction to HAProxy
278--------------------------
279
Davor Ocelic4094ce12017-12-19 23:30:39 +0100280HAProxy is written as "HAProxy" to designate the product, and as "haproxy" to
281designate the executable program, software package or a process. However, both
282are commonly used for both purposes, and are pronounced H-A-Proxy. Very early,
283"haproxy" used to stand for "high availability proxy" and the name was written
284in two separate words, though by now it means nothing else than "HAProxy".
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200285
286
Davor Ocelic4094ce12017-12-19 23:30:39 +01002873.1. What HAProxy is and isn't
288------------------------------
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200289
290HAProxy is :
291
292 - a TCP proxy : it can accept a TCP connection from a listening socket,
293 connect to a server and attach these sockets together allowing traffic to
Willy Tarreauec8962c2020-05-05 17:39:16 +0200294 flow in both directions; IPv4, IPv6 and even UNIX sockets are supported on
295 either side, so this can provide an easy way to translate addresses between
296 different families.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200297
298 - an HTTP reverse-proxy (called a "gateway" in HTTP terminology) : it presents
299 itself as a server, receives HTTP requests over connections accepted on a
300 listening TCP socket, and passes the requests from these connections to
Willy Tarreauec8962c2020-05-05 17:39:16 +0200301 servers using different connections. It may use any combination of HTTP/1.x
302 or HTTP/2 on any side and will even automatically detect the protocol
303 spoken on each side when ALPN is used over TLS.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200304
305 - an SSL terminator / initiator / offloader : SSL/TLS may be used on the
306 connection coming from the client, on the connection going to the server,
Willy Tarreauec8962c2020-05-05 17:39:16 +0200307 or even on both connections. A lot of settings can be applied per name
308 (SNI), and may be updated at runtime without restarting. Such setups are
309 extremely scalable and deployments involving tens to hundreds of thousands
310 of certificates were reported.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200311
312 - a TCP normalizer : since connections are locally terminated by the operating
313 system, there is no relation between both sides, so abnormal traffic such as
314 invalid packets, flag combinations, window advertisements, sequence numbers,
315 incomplete connections (SYN floods), or so will not be passed to the other
316 side. This protects fragile TCP stacks from protocol attacks, and also
317 allows to optimize the connection parameters with the client without having
318 to modify the servers' TCP stack settings.
319
320 - an HTTP normalizer : when configured to process HTTP traffic, only valid
321 complete requests are passed. This protects against a lot of protocol-based
322 attacks. Additionally, protocol deviations for which there is a tolerance
323 in the specification are fixed so that they don't cause problem on the
Davor Ocelic4094ce12017-12-19 23:30:39 +0100324 servers (e.g. multiple-line headers).
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200325
326 - an HTTP fixing tool : it can modify / fix / add / remove / rewrite the URL
327 or any request or response header. This helps fixing interoperability issues
328 in complex environments.
329
330 - a content-based switch : it can consider any element from the request to
331 decide what server to pass the request or connection to. Thus it is possible
Davor Ocelic4094ce12017-12-19 23:30:39 +0100332 to handle multiple protocols over a same port (e.g. HTTP, HTTPS, SSH).
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200333
334 - a server load balancer : it can load balance TCP connections and HTTP
335 requests. In TCP mode, load balancing decisions are taken for the whole
336 connection. In HTTP mode, decisions are taken per request.
337
338 - a traffic regulator : it can apply some rate limiting at various points,
339 protect the servers against overloading, adjust traffic priorities based on
340 the contents, and even pass such information to lower layers and outer
341 network components by marking packets.
342
343 - a protection against DDoS and service abuse : it can maintain a wide number
344 of statistics per IP address, URL, cookie, etc and detect when an abuse is
345 happening, then take action (slow down the offenders, block them, send them
346 to outdated contents, etc).
347
348 - an observation point for network troubleshooting : due to the precision of
349 the information reported in logs, it is often used to narrow down some
350 network-related issues.
351
352 - an HTTP compression offloader : it can compress responses which were not
353 compressed by the server, thus reducing the page load time for clients with
354 poor connectivity or using high-latency, mobile networks.
355
Willy Tarreauec8962c2020-05-05 17:39:16 +0200356 - a caching proxy : it may cache responses in RAM so that subsequent requests
357 for the same object avoid the cost of another network transfer from the
358 server as long as the object remains present and valid. It will however not
359 store objects to any persistent storage. Please note that this caching
360 feature is designed to be maintenance free and focuses solely on saving
361 haproxy's precious resources and not on save the server's resources. Caches
362 designed to optimize servers require much more tuning and flexibility. If
363 you instead need such an advanced cache, please use Varnish Cache, which
364 integrates perfectly with haproxy, especially when SSL/TLS is needed on any
365 side.
366
367 - a FastCGI gateway : FastCGI can be seen as a different representation of
368 HTTP, and as such, HAProxy can directly load-balance a farm comprising any
369 combination of FastCGI application servers without requiring to insert
370 another level of gateway between them. This results in resource savings and
371 a reduction of maintenance costs.
372
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200373HAProxy is not :
374
Davor Ocelic4094ce12017-12-19 23:30:39 +0100375 - an explicit HTTP proxy, i.e. the proxy that browsers use to reach the
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200376 internet. There are excellent open-source software dedicated for this task,
377 such as Squid. However HAProxy can be installed in front of such a proxy to
378 provide load balancing and high availability.
379
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200380 - a data scrubber : it will not modify the body of requests nor responses.
381
Willy Tarreauec8962c2020-05-05 17:39:16 +0200382 - a static web server : during startup, it isolates itself inside a chroot
383 jail and drops its privileges, so that it will not perform any single file-
384 system access once started. As such it cannot be turned into a static web
385 server (dynamic servers are supported through FastCGI however). There are
386 excellent open-source software for this such as Apache or Nginx, and
387 HAProxy can be easily installed in front of them to provide load balancing,
388 high availability and acceleration.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200389
390 - a packet-based load balancer : it will not see IP packets nor UDP datagrams,
391 will not perform NAT or even less DSR. These are tasks for lower layers.
392 Some kernel-based components such as IPVS (Linux Virtual Server) already do
393 this pretty well and complement perfectly with HAProxy.
394
395
3963.2. How HAProxy works
397----------------------
398
Willy Tarreauec8962c2020-05-05 17:39:16 +0200399HAProxy is an event-driven, non-blocking engine combining a very fast I/O layer
400with a priority-based, multi-threaded scheduler. As it is designed with a data
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200401forwarding goal in mind, its architecture is optimized to move data as fast as
Willy Tarreauec8962c2020-05-05 17:39:16 +0200402possible with the least possible operations. It focuses on optimizing the CPU
403cache's efficiency by sticking connections to the same CPU as long as possible.
404As such it implements a layered model offering bypass mechanisms at each level
405ensuring data doesn't reach higher levels unless needed. Most of the processing
406is performed in the kernel, and HAProxy does its best to help the kernel do the
407work as fast as possible by giving some hints or by avoiding certain operation
408when it guesses they could be grouped later. As a result, typical figures show
40915% of the processing time spent in HAProxy versus 85% in the kernel in TCP or
410HTTP close mode, and about 30% for HAProxy versus 70% for the kernel in HTTP
411keep-alive mode.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200412
413A single process can run many proxy instances; configurations as large as
Willy Tarreauec8962c2020-05-05 17:39:16 +0200414300000 distinct proxies in a single process were reported to run fine. A single
415core, single CPU setup is far more than enough for more than 99% users, and as
416such, users of containers and virtual machines are encouraged to use the
417absolute smallest images they can get to save on operational costs and simplify
418troubleshooting. However the machine HAProxy runs on must never ever swap, and
419its CPU must not be artificially throttled (sub-CPU allocation in hypervisors)
420nor be shared with compute-intensive processes which would induce a very high
421context-switch latency.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200422
Willy Tarreauec8962c2020-05-05 17:39:16 +0200423Threading allows to exploit all available processing capacity by using one
424thread per CPU core. This is mostly useful for SSL or when data forwarding
425rates above 40 Gbps are needed. In such cases it is critically important to
426avoid communications between multiple physical CPUs, which can cause strong
427bottlenecks in the network stack and in HAProxy itself. While counter-intuitive
428to some, the first thing to do when facing some performance issues is often to
429reduce the number of CPUs HAProxy runs on.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200430
431HAProxy only requires the haproxy executable and a configuration file to run.
432For logging it is highly recommended to have a properly configured syslog daemon
Willy Tarreauec8962c2020-05-05 17:39:16 +0200433and log rotations in place. Logs may also be sent to stdout/stderr, which can be
434useful inside containers. The configuration files are parsed before starting,
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200435then HAProxy tries to bind all listening sockets, and refuses to start if
436anything fails. Past this point it cannot fail anymore. This means that there
437are no runtime failures and that if it accepts to start, it will work until it
438is stopped.
439
440Once HAProxy is started, it does exactly 3 things :
441
442 - process incoming connections;
443
444 - periodically check the servers' status (known as health checks);
445
446 - exchange information with other haproxy nodes.
447
448Processing incoming connections is by far the most complex task as it depends
449on a lot of configuration possibilities, but it can be summarized as the 9 steps
450below :
451
452 - accept incoming connections from listening sockets that belong to a
453 configuration entity known as a "frontend", which references one or multiple
454 listening addresses;
455
456 - apply the frontend-specific processing rules to these connections that may
457 result in blocking them, modifying some headers, or intercepting them to
458 execute some internal applets such as the statistics page or the CLI;
459
460 - pass these incoming connections to another configuration entity representing
461 a server farm known as a "backend", which contains the list of servers and
462 the load balancing strategy for this server farm;
463
464 - apply the backend-specific processing rules to these connections;
465
466 - decide which server to forward the connection to according to the load
467 balancing strategy;
468
469 - apply the backend-specific processing rules to the response data;
470
471 - apply the frontend-specific processing rules to the response data;
472
473 - emit a log to report what happened in fine details;
474
475 - in HTTP, loop back to the second step to wait for a new request, otherwise
476 close the connection.
477
478Frontends and backends are sometimes considered as half-proxies, since they only
479look at one side of an end-to-end connection; the frontend only cares about the
480clients while the backend only cares about the servers. HAProxy also supports
481full proxies which are exactly the union of a frontend and a backend. When HTTP
482processing is desired, the configuration will generally be split into frontends
483and backends as they open a lot of possibilities since any frontend may pass a
484connection to any backend. With TCP-only proxies, using frontends and backends
485rarely provides a benefit and the configuration can be more readable with full
486proxies.
487
488
4893.3. Basic features
490-------------------
491
492This section will enumerate a number of features that HAProxy implements, some
493of which are generally expected from any modern load balancer, and some of
494which are a direct benefit of HAProxy's architecture. More advanced features
495will be detailed in the next section.
496
497
4983.3.1. Basic features : Proxying
499--------------------------------
500
501Proxying is the action of transferring data between a client and a server over
Patrick Starrdce734e2017-10-09 13:17:12 +0700502two independent connections. The following basic features are supported by
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200503HAProxy regarding proxying and connection management :
504
505 - Provide the server with a clean connection to protect them against any
506 client-side defect or attack;
507
Patrick Starrdce734e2017-10-09 13:17:12 +0700508 - Listen to multiple IP addresses and/or ports, even port ranges;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200509
Davor Ocelic4094ce12017-12-19 23:30:39 +0100510 - Transparent accept : intercept traffic targeting any arbitrary IP address
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200511 that doesn't even belong to the local system;
512
513 - Server port doesn't need to be related to listening port, and may even be
514 translated by a fixed offset (useful with ranges);
515
516 - Transparent connect : spoof the client's (or any) IP address if needed
517 when connecting to the server;
518
519 - Provide a reliable return IP address to the servers in multi-site LBs;
520
521 - Offload the server thanks to buffers and possibly short-lived connections
522 to reduce their concurrent connection count and their memory footprint;
523
Davor Ocelic4094ce12017-12-19 23:30:39 +0100524 - Optimize TCP stacks (e.g. SACK), congestion control, and reduce RTT impacts;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200525
Davor Ocelic4094ce12017-12-19 23:30:39 +0100526 - Support different protocol families on both sides (e.g. IPv4/IPv6/Unix);
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200527
528 - Timeout enforcement : HAProxy supports multiple levels of timeouts depending
529 on the stage the connection is, so that a dead client or server, or an
530 attacker cannot be granted resources for too long;
531
532 - Protocol validation: HTTP, SSL, or payload are inspected and invalid
533 protocol elements are rejected, unless instructed to accept them anyway;
534
535 - Policy enforcement : ensure that only what is allowed may be forwarded;
536
537 - Both incoming and outgoing connections may be limited to certain network
538 namespaces (Linux only), making it easy to build a cross-container,
539 multi-tenant load balancer;
540
541 - PROXY protocol presents the client's IP address to the server even for
542 non-HTTP traffic. This is an HAProxy extension that was adopted by a number
543 of third-party products by now, at least these ones at the time of writing :
544 - client : haproxy, stud, stunnel, exaproxy, ELB, squid
545 - server : haproxy, stud, postfix, exim, nginx, squid, node.js, varnish
546
547
5483.3.2. Basic features : SSL
549---------------------------
550
551HAProxy's SSL stack is recognized as one of the most featureful according to
552Google's engineers (http://istlsfastyet.com/). The most commonly used features
553making it quite complete are :
554
555 - SNI-based multi-hosting with no limit on sites count and focus on
556 performance. At least one deployment is known for running 50000 domains
557 with their respective certificates;
558
559 - support for wildcard certificates reduces the need for many certificates ;
560
561 - certificate-based client authentication with configurable policies on
562 failure to present a valid certificate. This allows to present a different
563 server farm to regenerate the client certificate for example;
564
565 - authentication of the backend server ensures the backend server is the real
566 one and not a man in the middle;
567
Patrick Starrdce734e2017-10-09 13:17:12 +0700568 - authentication with the backend server lets the backend server know it's
569 really the expected haproxy node that is connecting to it;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200570
571 - TLS NPN and ALPN extensions make it possible to reliably offload SPDY/HTTP2
572 connections and pass them in clear text to backend servers;
573
574 - OCSP stapling further reduces first page load time by delivering inline an
575 OCSP response when the client requests a Certificate Status Request;
576
577 - Dynamic record sizing provides both high performance and low latency, and
578 significantly reduces page load time by letting the browser start to fetch
579 new objects while packets are still in flight;
580
581 - permanent access to all relevant SSL/TLS layer information for logging,
Davor Ocelic4094ce12017-12-19 23:30:39 +0100582 access control, reporting etc. These elements can be embedded into HTTP
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200583 header or even as a PROXY protocol extension so that the offloaded server
584 gets all the information it would have had if it performed the SSL
585 termination itself.
586
587 - Detect, log and block certain known attacks even on vulnerable SSL libs,
588 such as the Heartbleed attack affecting certain versions of OpenSSL.
589
Pavlos Parissisba56d9c2015-08-24 13:14:32 +0200590 - support for stateless session resumption (RFC 5077 TLS Ticket extension).
591 TLS tickets can be updated from CLI which provides them means to implement
592 Perfect Forward Secrecy by frequently rotating the tickets.
593
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200594
5953.3.3. Basic features : Monitoring
596----------------------------------
597
598HAProxy focuses a lot on availability. As such it cares about servers state,
599and about reporting its own state to other network components :
600
Patrick Starrdce734e2017-10-09 13:17:12 +0700601 - Servers' state is continuously monitored using per-server parameters. This
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200602 ensures the path to the server is operational for regular traffic;
603
604 - Health checks support two hysteresis for up and down transitions in order
605 to protect against state flapping;
606
607 - Checks can be sent to a different address/port/protocol : this makes it
608 easy to check a single service that is considered representative of multiple
609 ones, for example the HTTPS port for an HTTP+HTTPS server.
610
611 - Servers can track other servers and go down simultaneously : this ensures
Davor Ocelic4094ce12017-12-19 23:30:39 +0100612 that servers hosting multiple services can fail atomically and that no one
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200613 will be sent to a partially failed server;
614
615 - Agents may be deployed on the server to monitor load and health : a server
616 may be interested in reporting its load, operational status, administrative
Patrick Starrdce734e2017-10-09 13:17:12 +0700617 status independently from what health checks can see. By running a simple
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200618 agent on the server, it's possible to consider the server's view of its own
619 health in addition to the health checks validating the whole path;
620
621 - Various check methods are available : TCP connect, HTTP request, SMTP hello,
622 SSL hello, LDAP, SQL, Redis, send/expect scripts, all with/without SSL;
623
624 - State change is notified in the logs and stats page with the failure reason
Davor Ocelic4094ce12017-12-19 23:30:39 +0100625 (e.g. the HTTP response received at the moment the failure was detected). An
Willy Tarreaueff04f42015-08-27 14:44:43 +0200626 e-mail can also be sent to a configurable address upon such a change ;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200627
628 - Server state is also reported on the stats interface and can be used to take
629 routing decisions so that traffic may be sent to different farms depending
Davor Ocelic4094ce12017-12-19 23:30:39 +0100630 on their sizes and/or health (e.g. loss of an inter-DC link);
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200631
632 - HAProxy can use health check requests to pass information to the servers,
Davor Ocelic4094ce12017-12-19 23:30:39 +0100633 such as their names, weight, the number of other servers in the farm etc.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200634 so that servers can adjust their response and decisions based on this
Davor Ocelic4094ce12017-12-19 23:30:39 +0100635 knowledge (e.g. postpone backups to keep more CPU available);
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200636
637 - Servers can use health checks to report more detailed state than just on/off
Davor Ocelic4094ce12017-12-19 23:30:39 +0100638 (e.g. I would like to stop, please stop sending new visitors);
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200639
640 - HAProxy itself can report its state to external components such as routers
641 or other load balancers, allowing to build very complete multi-path and
642 multi-layer infrastructures.
643
644
6453.3.4. Basic features : High availability
646-----------------------------------------
647
648Just like any serious load balancer, HAProxy cares a lot about availability to
649ensure the best global service continuity :
650
Davor Ocelic4094ce12017-12-19 23:30:39 +0100651 - Only valid servers are used ; the other ones are automatically evicted from
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200652 load balancing farms ; under certain conditions it is still possible to
653 force to use them though;
654
655 - Support for a graceful shutdown so that it is possible to take servers out
656 of a farm without affecting any connection;
657
658 - Backup servers are automatically used when active servers are down and
659 replace them so that sessions are not lost when possible. This also allows
Davor Ocelic4094ce12017-12-19 23:30:39 +0100660 to build multiple paths to reach the same server (e.g. multiple interfaces);
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200661
662 - Ability to return a global failed status for a farm when too many servers
663 are down. This, combined with the monitoring capabilities makes it possible
664 for an upstream component to choose a different LB node for a given service;
665
666 - Stateless design makes it easy to build clusters : by design, HAProxy does
667 its best to ensure the highest service continuity without having to store
668 information that could be lost in the event of a failure. This ensures that
669 a takeover is the most seamless possible;
670
671 - Integrates well with standard VRRP daemon keepalived : HAProxy easily tells
Patrick Starrdce734e2017-10-09 13:17:12 +0700672 keepalived about its state and copes very well with floating virtual IP
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200673 addresses. Note: only use IP redundancy protocols (VRRP/CARP) over cluster-
674 based solutions (Heartbeat, ...) as they're the ones offering the fastest,
675 most seamless, and most reliable switchover.
676
677
6783.3.5. Basic features : Load balancing
679--------------------------------------
680
681HAProxy offers a fairly complete set of load balancing features, most of which
682are unfortunately not available in a number of other load balancing products :
683
Willy Tarreauec8962c2020-05-05 17:39:16 +0200684 - no less than 10 load balancing algorithms are supported, some of which apply
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200685 to input data to offer an infinite list of possibilities. The most common
686 ones are round-robin (for short connections, pick each server in turn),
687 leastconn (for long connections, pick the least recently used of the servers
688 with the lowest connection count), source (for SSL farms or terminal server
Davor Ocelic4094ce12017-12-19 23:30:39 +0100689 farms, the server directly depends on the client's source address), URI (for
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200690 HTTP caches, the server directly depends on the HTTP URI), hdr (the server
691 directly depends on the contents of a specific HTTP header field), first
692 (for short-lived virtual machines, all connections are packed on the
693 smallest possible subset of servers so that unused ones can be powered
694 down);
695
696 - all algorithms above support per-server weights so that it is possible to
697 accommodate from different server generations in a farm, or direct a small
698 fraction of the traffic to specific servers (debug mode, running the next
699 version of the software, etc);
700
701 - dynamic weights are supported for round-robin, leastconn and consistent
702 hashing ; this allows server weights to be modified on the fly from the CLI
703 or even by an agent running on the server;
704
705 - slow-start is supported whenever a dynamic weight is supported; this allows
706 a server to progressively take the traffic. This is an important feature
707 for fragile application servers which require to compile classes at runtime
708 as well as cold caches which need to fill up before being run at full
709 throttle;
710
711 - hashing can apply to various elements such as client's source address, URL
712 components, query string element, header field values, POST parameter, RDP
713 cookie;
714
715 - consistent hashing protects server farms against massive redistribution when
716 adding or removing servers in a farm. That's very important in large cache
717 farms and it allows slow-start to be used to refill cold caches;
718
719 - a number of internal metrics such as the number of connections per server,
720 per backend, the amount of available connection slots in a backend etc makes
721 it possible to build very advanced load balancing strategies.
722
723
7243.3.6. Basic features : Stickiness
725----------------------------------
726
727Application load balancing would be useless without stickiness. HAProxy provides
728a fairly comprehensive set of possibilities to maintain a visitor on the same
729server even across various events such as server addition/removal, down/up
730cycles, and some methods are designed to be resistant to the distance between
731multiple load balancing nodes in that they don't require any replication :
732
733 - stickiness information can be individually matched and learned from
734 different places if desired. For example a JSESSIONID cookie may be matched
735 both in a cookie and in the URL. Up to 8 parallel sources can be learned at
736 the same time and each of them may point to a different stick-table;
737
738 - stickiness information can come from anything that can be seen within a
739 request or response, including source address, TCP payload offset and
Patrick Starrdce734e2017-10-09 13:17:12 +0700740 length, HTTP query string elements, header field values, cookies, and so
Davor Ocelic4094ce12017-12-19 23:30:39 +0100741 on.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200742
Davor Ocelic4094ce12017-12-19 23:30:39 +0100743 - stick-tables are replicated between all nodes in a multi-master fashion;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200744
745 - commonly used elements such as SSL-ID or RDP cookies (for TSE farms) are
746 directly accessible to ease manipulation;
747
Davor Ocelic4094ce12017-12-19 23:30:39 +0100748 - all sticking rules may be dynamically conditioned by ACLs;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200749
750 - it is possible to decide not to stick to certain servers, such as backup
751 servers, so that when the nominal server comes back, it automatically takes
752 the load back. This is often used in multi-path environments;
753
Davor Ocelic4094ce12017-12-19 23:30:39 +0100754 - in HTTP it is often preferred not to learn anything and instead manipulate
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200755 a cookie dedicated to stickiness. For this, it's possible to detect,
756 rewrite, insert or prefix such a cookie to let the client remember what
757 server was assigned;
758
759 - the server may decide to change or clean the stickiness cookie on logout,
760 so that leaving visitors are automatically unbound from the server;
761
762 - using ACL-based rules it is also possible to selectively ignore or enforce
763 stickiness regardless of the server's state; combined with advanced health
764 checks, that helps admins verify that the server they're installing is up
765 and running before presenting it to the whole world;
766
767 - an innovative mechanism to set a maximum idle time and duration on cookies
768 ensures that stickiness can be smoothly stopped on devices which are never
769 closed (smartphones, TVs, home appliances) without having to store them on
770 persistent storage;
771
772 - multiple server entries may share the same stickiness keys so that
773 stickiness is not lost in multi-path environments when one path goes down;
774
775 - soft-stop ensures that only users with stickiness information will continue
776 to reach the server they've been assigned to but no new users will go there.
777
778
7793.3.7. Basic features : Sampling and converting information
780-----------------------------------------------------------
781
782HAProxy supports information sampling using a wide set of "sample fetch
783functions". The principle is to extract pieces of information known as samples,
784for immediate use. This is used for stickiness, to build conditions, to produce
785information in logs or to enrich HTTP headers.
786
787Samples can be fetched from various sources :
788
789 - constants : integers, strings, IP addresses, binary blocks;
790
791 - the process : date, environment variables, server/frontend/backend/process
792 state, byte/connection counts/rates, queue length, random generator, ...
793
794 - variables : per-session, per-request, per-response variables;
795
796 - the client connection : source and destination addresses and ports, and all
797 related statistics counters;
798
799 - the SSL client session : protocol, version, algorithm, cipher, key size,
800 session ID, all client and server certificate fields, certificate serial,
801 SNI, ALPN, NPN, client support for certain extensions;
802
803 - request and response buffers contents : arbitrary payload at offset/length,
804 data length, RDP cookie, decoding of SSL hello type, decoding of TLS SNI;
805
806 - HTTP (request and response) : method, URI, path, query string arguments,
Davor Ocelic4094ce12017-12-19 23:30:39 +0100807 status code, headers values, positional header value, cookies, captures,
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200808 authentication, body elements;
809
810A sample may then pass through a number of operators known as "converters" to
811experience some transformation. A converter consumes a sample and produces a
812new one, possibly of a completely different type. For example, a converter may
813be used to return only the integer length of the input string, or could turn a
814string to upper case. Any arbitrary number of converters may be applied in
815series to a sample before final use. Among all available sample converters, the
816following ones are the most commonly used :
817
818 - arithmetic and logic operators : they make it possible to perform advanced
819 computation on input data, such as computing ratios, percentages or simply
820 converting from one unit to another one;
821
Patrick Starrdce734e2017-10-09 13:17:12 +0700822 - IP address masks are useful when some addresses need to be grouped by larger
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200823 networks;
824
Davor Ocelic4094ce12017-12-19 23:30:39 +0100825 - data representation : URL-decode, base64, hex, JSON strings, hashing;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200826
827 - string conversion : extract substrings at fixed positions, fixed length,
828 extract specific fields around certain delimiters, extract certain words,
Patrick Starrdce734e2017-10-09 13:17:12 +0700829 change case, apply regex-based substitution;
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200830
Davor Ocelic4094ce12017-12-19 23:30:39 +0100831 - date conversion : convert to HTTP date format, convert local to UTC and
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200832 conversely, add or remove offset;
833
834 - lookup an entry in a stick table to find statistics or assigned server;
835
836 - map-based key-to-value conversion from a file (mostly used for geolocation).
837
838
8393.3.8. Basic features : Maps
840----------------------------
841
842Maps are a powerful type of converter consisting in loading a two-columns file
843into memory at boot time, then looking up each input sample from the first
844column and either returning the corresponding pattern on the second column if
845the entry was found, or returning a default value. The output information also
846being a sample, it can in turn experience other transformations including other
847map lookups. Maps are most commonly used to translate the client's IP address
848to an AS number or country code since they support a longest match for network
849addresses but they can be used for various other purposes.
850
851Part of their strength comes from being updatable on the fly either from the CLI
852or from certain actions using other samples, making them capable of storing and
853retrieving information between subsequent accesses. Another strength comes from
Patrick Starrdce734e2017-10-09 13:17:12 +0700854the binary tree based indexation which makes them extremely fast even when they
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200855contain hundreds of thousands of entries, making geolocation very cheap and easy
856to set up.
857
858
8593.3.9. Basic features : ACLs and conditions
860-------------------------------------------
861
862Most operations in HAProxy can be made conditional. Conditions are built by
863combining multiple ACLs using logic operators (AND, OR, NOT). Each ACL is a
864series of tests based on the following elements :
865
866 - a sample fetch method to retrieve the element to test ;
867
868 - an optional series of converters to transform the element ;
869
870 - a list of patterns to match against ;
871
872 - a matching method to indicate how to compare the patterns with the sample
873
874For example, the sample may be taken from the HTTP "Host" header, it could then
875be converted to lower case, then matched against a number of regex patterns
876using the regex matching method.
877
878Technically, ACLs are built on the same core as the maps, they share the exact
879same internal structure, pattern matching methods and performance. The only real
880difference is that instead of returning a sample, they only return "found" or
881or "not found". In terms of usage, ACL patterns may be declared inline in the
882configuration file and do not require their own file. ACLs may be named for ease
883of use or to make configurations understandable. A named ACL may be declared
884multiple times and it will evaluate all definitions in turn until one matches.
885
886About 13 different pattern matching methods are provided, among which IP address
887mask, integer ranges, substrings, regex. They work like functions, and just like
888with any programming language, only what is needed is evaluated, so when a
889condition involving an OR is already true, next ones are not evaluated, and
890similarly when a condition involving an AND is already false, the rest of the
891condition is not evaluated.
892
893There is no practical limit to the number of declared ACLs, and a handful of
894commonly used ones are provided. However experience has shown that setups using
895a lot of named ACLs are quite hard to troubleshoot and that sometimes using
Patrick Starrdce734e2017-10-09 13:17:12 +0700896anonymous ACLs inline is easier as it requires less references out of the scope
Davor Ocelic4094ce12017-12-19 23:30:39 +0100897being analyzed.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200898
899
9003.3.10. Basic features : Content switching
901------------------------------------------
902
903HAProxy implements a mechanism known as content-based switching. The principle
904is that a connection or request arrives on a frontend, then the information
905carried with this request or connection are processed, and at this point it is
906possible to write ACLs-based conditions making use of these information to
907decide what backend will process the request. Thus the traffic is directed to
908one backend or another based on the request's contents. The most common example
909consists in using the Host header and/or elements from the path (sub-directories
910or file-name extensions) to decide whether an HTTP request targets a static
911object or the application, and to route static objects traffic to a backend made
912of fast and light servers, and all the remaining traffic to a more complex
913application server, thus constituting a fine-grained virtual hosting solution.
914This is quite convenient to make multiple technologies coexist as a more global
915solution.
916
917Another use case of content-switching consists in using different load balancing
918algorithms depending on various criteria. A cache may use a URI hash while an
Davor Ocelic4094ce12017-12-19 23:30:39 +0100919application would use round-robin.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200920
921Last but not least, it allows multiple customers to use a small share of a
922common resource by enforcing per-backend (thus per-customer connection limits).
923
924Content switching rules scale very well, though their performance may depend on
925the number and complexity of the ACLs in use. But it is also possible to write
926dynamic content switching rules where a sample value directly turns into a
927backend name and without making use of ACLs at all. Such configurations have
928been reported to work fine at least with 300000 backends in production.
929
930
9313.3.11. Basic features : Stick-tables
932-------------------------------------
933
934Stick-tables are commonly used to store stickiness information, that is, to keep
935a reference to the server a certain visitor was directed to. The key is then the
936identifier associated with the visitor (its source address, the SSL ID of the
937connection, an HTTP or RDP cookie, the customer number extracted from the URL or
938from the payload, ...) and the stored value is then the server's identifier.
939
940Stick tables may use 3 different types of samples for their keys : integers,
941strings and addresses. Only one stick-table may be referenced in a proxy, and it
Davor Ocelic4094ce12017-12-19 23:30:39 +0100942is designated everywhere with the proxy name. Up to 8 keys may be tracked in
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200943parallel. The server identifier is committed during request or response
944processing once both the key and the server are known.
945
946Stick-table contents may be replicated in active-active mode with other HAProxy
947nodes known as "peers" as well as with the new process during a reload operation
948so that all load balancing nodes share the same information and take the same
Davor Ocelic4094ce12017-12-19 23:30:39 +0100949routing decision if client's requests are spread over multiple nodes.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200950
951Since stick-tables are indexed on what allows to recognize a client, they are
952often also used to store extra information such as per-client statistics. The
953extra statistics take some extra space and need to be explicitly declared. The
954type of statistics that may be stored includes the input and output bandwidth,
955the number of concurrent connections, the connection rate and count over a
956period, the amount and frequency of errors, some specific tags and counters,
Davor Ocelic4094ce12017-12-19 23:30:39 +0100957etc. In order to support keeping such information without being forced to
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200958stick to a given server, a special "tracking" feature is implemented and allows
959to track up to 3 simultaneous keys from different tables at the same time
960regardless of stickiness rules. Each stored statistics may be searched, dumped
961and cleared from the CLI and adds to the live troubleshooting capabilities.
962
963While this mechanism can be used to surclass a returning visitor or to adjust
Davor Ocelic4094ce12017-12-19 23:30:39 +0100964the delivered quality of service depending on good or bad behavior, it is
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200965mostly used to fight against service abuse and more generally DDoS as it allows
Davor Ocelic4094ce12017-12-19 23:30:39 +0100966to build complex models to detect certain bad behaviors at a high processing
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200967speed.
968
969
Davor Ocelic4094ce12017-12-19 23:30:39 +01009703.3.12. Basic features : Formatted strings
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200971-----------------------------------------
972
973There are many places where HAProxy needs to manipulate character strings, such
974as logs, redirects, header additions, and so on. In order to provide the
Davor Ocelic4094ce12017-12-19 23:30:39 +0100975greatest flexibility, the notion of Formatted strings was introduced, initially
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200976for logging purposes, which explains why it's still called "log-format". These
977strings contain escape characters allowing to introduce various dynamic data
978including variables and sample fetch expressions into strings, and even to
979adjust the encoding while the result is being turned into a string (for example,
Willy Tarreauec8962c2020-05-05 17:39:16 +0200980adding quotes). This provides a powerful way to build header contents, to build
981response data or even response templates, or to customize log lines.
982Additionally, in order to remain simple to build most common strings, about 50
983special tags are provided as shortcuts for information commonly used in logs.
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200984
985
9863.3.13. Basic features : HTTP rewriting and redirection
987-------------------------------------------------------
988
989Installing a load balancer in front of an application that was never designed
990for this can be a challenging task without the proper tools. One of the most
991commonly requested operation in this case is to adjust requests and response
992headers to make the load balancer appear as the origin server and to fix hard
993coded information. This comes with changing the path in requests (which is
994strongly advised against), modifying Host header field, modifying the Location
995response header field for redirects, modifying the path and domain attribute
996for cookies, and so on. It also happens that a number of servers are somewhat
997verbose and tend to leak too much information in the response, making them more
Davor Ocelic4094ce12017-12-19 23:30:39 +0100998vulnerable to targeted attacks. While it's theoretically not the role of a load
Willy Tarreaud8e42b62015-08-18 21:51:36 +0200999balancer to clean this up, in practice it's located at the best place in the
1000infrastructure to guarantee that everything is cleaned up.
1001
1002Similarly, sometimes the load balancer will have to intercept some requests and
1003respond with a redirect to a new target URL. While some people tend to confuse
1004redirects and rewriting, these are two completely different concepts, since the
1005rewriting makes the client and the server see different things (and disagree on
1006the location of the page being visited) while redirects ask the client to visit
1007the new URL so that it sees the same location as the server.
1008
1009In order to do this, HAProxy supports various possibilities for rewriting and
Davor Ocelic4094ce12017-12-19 23:30:39 +01001010redirects, among which :
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001011
1012 - regex-based URL and header rewriting in requests and responses. Regex are
1013 the most commonly used tool to modify header values since they're easy to
1014 manipulate and well understood;
1015
Davor Ocelic4094ce12017-12-19 23:30:39 +01001016 - headers may also be appended, deleted or replaced based on formatted strings
1017 so that it is possible to pass information there (e.g. client side TLS
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001018 algorithm and cipher);
1019
1020 - HTTP redirects can use any 3xx code to a relative, absolute, or completely
Davor Ocelic4094ce12017-12-19 23:30:39 +01001021 dynamic (formatted string) URI;
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001022
1023 - HTTP redirects also support some extra options such as setting or clearing
1024 a specific cookie, dropping the query string, appending a slash if missing,
1025 and so on;
1026
Willy Tarreauec8962c2020-05-05 17:39:16 +02001027 - a powerful "return" directive allows to customize every part of a response
1028 like status, headers, body using dynamic contents or even template files.
1029
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001030 - all operations support ACL-based conditions;
1031
1032
10333.3.14. Basic features : Server protection
1034------------------------------------------
1035
Davor Ocelic4094ce12017-12-19 23:30:39 +01001036HAProxy does a lot to maximize service availability, and for this it takes
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001037large efforts to protect servers against overloading and attacks. The first
1038and most important point is that only complete and valid requests are forwarded
1039to the servers. The initial reason is that HAProxy needs to find the protocol
1040elements it needs to stay synchronized with the byte stream, and the second
1041reason is that until the request is complete, there is no way to know if some
1042elements will change its semantics. The direct benefit from this is that servers
1043are not exposed to invalid or incomplete requests. This is a very effective
1044protection against slowloris attacks, which have almost no impact on HAProxy.
1045
1046Another important point is that HAProxy contains buffers to store requests and
1047responses, and that by only sending a request to a server when it's complete and
1048by reading the whole response very quickly from the local network, the server
1049side connection is used for a very short time and this preserves server
1050resources as much as possible.
1051
1052A direct extension to this is that HAProxy can artificially limit the number of
1053concurrent connections or outstanding requests to a server, which guarantees
1054that the server will never be overloaded even if it continuously runs at 100% of
1055its capacity during traffic spikes. All excess requests will simply be queued to
1056be processed when one slot is released. In the end, this huge resource savings
1057most often ensures so much better server response times that it ends up actually
1058being faster than by overloading the server. Queued requests may be redispatched
1059to other servers, or even aborted in queue when the client aborts, which also
1060protects the servers against the "reload effect", where each click on "reload"
1061by a visitor on a slow-loading page usually induces a new request and maintains
1062the server in an overloaded state.
1063
1064The slow-start mechanism also protects restarting servers against high traffic
1065levels while they're still finalizing their startup or compiling some classes.
1066
1067Regarding the protocol-level protection, it is possible to relax the HTTP parser
Davor Ocelic4094ce12017-12-19 23:30:39 +01001068to accept non standard-compliant but harmless requests or responses and even to
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001069fix them. This allows bogus applications to be accessible while a fix is being
Patrick Starrdce734e2017-10-09 13:17:12 +07001070developed. In parallel, offending messages are completely captured with a
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001071detailed report that help developers spot the issue in the application. The most
1072dangerous protocol violations are properly detected and dealt with and fixed.
1073For example malformed requests or responses with two Content-length headers are
1074either fixed if the values are exactly the same, or rejected if they differ,
1075since it becomes a security problem. Protocol inspection is not limited to HTTP,
1076it is also available for other protocols like TLS or RDP.
1077
1078When a protocol violation or attack is detected, there are various options to
1079respond to the user, such as returning the common "HTTP 400 bad request",
Davor Ocelic4094ce12017-12-19 23:30:39 +01001080closing the connection with a TCP reset, or faking an error after a long delay
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001081("tarpit") to confuse the attacker. All of these contribute to protecting the
1082servers by discouraging the offending client from pursuing an attack that
1083becomes very expensive to maintain.
1084
1085HAProxy also proposes some more advanced options to protect against accidental
1086data leaks and session crossing. Not only it can log suspicious server responses
1087but it will also log and optionally block a response which might affect a given
1088visitors' confidentiality. One such example is a cacheable cookie appearing in a
1089cacheable response and which may result in an intermediary cache to deliver it
1090to another visitor, causing an accidental session sharing.
1091
1092
10933.3.15. Basic features : Logging
1094--------------------------------
1095
1096Logging is an extremely important feature for a load balancer, first because a
Davor Ocelic4094ce12017-12-19 23:30:39 +01001097load balancer is often wrongly accused of causing the problems it reveals, and
1098second because it is placed at a critical point in an infrastructure where all
1099normal and abnormal activity needs to be analyzed and correlated with other
1100components.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001101
1102HAProxy provides very detailed logs, with millisecond accuracy and the exact
Davor Ocelic4094ce12017-12-19 23:30:39 +01001103connection accept time that can be searched in firewalls logs (e.g. for NAT
Alain Belkadiac520952019-07-05 10:12:40 +02001104correlation). By default, TCP and HTTP logs are quite detailed and contain
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001105everything needed for troubleshooting, such as source IP address and port,
1106frontend, backend, server, timers (request receipt duration, queue duration,
1107connection setup time, response headers time, data transfer time), global
1108process state, connection counts, queue status, retries count, detailed
1109stickiness actions and disconnect reasons, header captures with a safe output
1110encoding. It is then possible to extend or replace this format to include any
1111sampled data, variables, captures, resulting in very detailed information. For
Davor Ocelic4094ce12017-12-19 23:30:39 +01001112example it is possible to log the number of cumulative requests or number of
1113different URLs visited by a client.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001114
1115The log level may be adjusted per request using standard ACLs, so it is possible
1116to automatically silent some logs considered as pollution and instead raise
Davor Ocelic4094ce12017-12-19 23:30:39 +01001117warnings when some abnormal behavior happen for a small part of the traffic
1118(e.g. too many URLs or HTTP errors for a source address). Administrative logs
1119are also emitted with their own levels to inform about the loss or recovery of a
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001120server for example.
1121
Davor Ocelic4094ce12017-12-19 23:30:39 +01001122Each frontend and backend may use multiple independent log outputs, which eases
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001123multi-tenancy. Logs are preferably sent over UDP, maybe JSON-encoded, and are
Willy Tarreauec8962c2020-05-05 17:39:16 +02001124truncated after a configurable line length in order to guarantee delivery. But
1125it is also possible to sned them to stdout/stderr or any file descriptor, as
1126well as to a ring buffer that a client can subscribe to in order to retrieve
1127them.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001128
1129
11303.3.16. Basic features : Statistics
1131-----------------------------------
1132
1133HAProxy provides a web-based statistics reporting interface with authentication,
1134security levels and scopes. It is thus possible to provide each hosted customer
1135with his own page showing only his own instances. This page can be located in a
1136hidden URL part of the regular web site so that no new port needs to be opened.
1137This page may also report the availability of other HAProxy nodes so that it is
1138easy to spot if everything works as expected at a glance. The view is synthetic
1139with a lot of details accessible (such as error causes, last access and last
1140change duration, etc), which are also accessible as a CSV table that other tools
1141may import to draw graphs. The page may self-refresh to be used as a monitoring
1142page on a large display. In administration mode, the page also allows to change
1143server state to ease maintenance operations.
1144
Willy Tarreauec8962c2020-05-05 17:39:16 +02001145A Prometheus exporter is also provided so that the statistics can be consumed
1146in a different format depending on the deployment.
1147
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001148
11493.4. Advanced features
1150----------------------
1151
11523.4.1. Advanced features : Management
1153-------------------------------------
1154
1155HAProxy is designed to remain extremely stable and safe to manage in a regular
1156production environment. It is provided as a single executable file which doesn't
1157require any installation process. Multiple versions can easily coexist, meaning
1158that it's possible (and recommended) to upgrade instances progressively by
Davor Ocelic4094ce12017-12-19 23:30:39 +01001159order of importance instead of migrating all of them at once. Configuration
1160files are easily versioned. Configuration checking is done off-line so it
1161doesn't require to restart a service that will possibly fail. During
1162configuration checks, a number of advanced mistakes may be detected (e.g. a rule
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001163hiding another one, or stickiness that will not work) and detailed warnings and
1164configuration hints are proposed to fix them. Backwards configuration file
1165compatibility goes very far away in time, with version 1.5 still fully
1166supporting configurations for versions 1.1 written 13 years before, and 1.6
1167only dropping support for almost unused, obsolete keywords that can be done
1168differently. The configuration and software upgrade mechanism is smooth and non
1169disruptive in that it allows old and new processes to coexist on the system,
Davor Ocelic4094ce12017-12-19 23:30:39 +01001170each handling its own connections. System status, build options, and library
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001171compatibility are reported on startup.
1172
1173Some advanced features allow an application administrator to smoothly stop a
1174server, detect when there's no activity on it anymore, then take it off-line,
1175stop it, upgrade it and ensure it doesn't take any traffic while being upgraded,
1176then test it again through the normal path without opening it to the public, and
1177all of this without touching HAProxy at all. This ensures that even complicated
1178production operations may be done during opening hours with all technical
1179resources available.
1180
1181The process tries to save resources as much as possible, uses memory pools to
1182save on allocation time and limit memory fragmentation, releases payload buffers
1183as soon as their contents are sent, and supports enforcing strong memory limits
1184above which connections have to wait for a buffer to become available instead of
1185allocating more memory. This system helps guarantee memory usage in certain
1186strict environments.
1187
1188A command line interface (CLI) is available as a UNIX or TCP socket, to perform
1189a number of operations and to retrieve troubleshooting information. Everything
1190done on this socket doesn't require a configuration change, so it is mostly used
1191for temporary changes. Using this interface it is possible to change a server's
1192address, weight and status, to consult statistics and clear counters, dump and
1193clear stickiness tables, possibly selectively by key criteria, dump and kill
1194client-side and server-side connections, dump captured errors with a detailed
1195analysis of the exact cause and location of the error, dump, add and remove
1196entries from ACLs and maps, update TLS shared secrets, apply connection limits
1197and rate limits on the fly to arbitrary frontends (useful in shared hosting
1198environments), and disable a specific frontend to release a listening port
1199(useful when daytime operations are forbidden and a fix is needed nonetheless).
Willy Tarreauec8962c2020-05-05 17:39:16 +02001200Updating certificates and their configuration on the fly is permitted, as well
1201as enabling and consulting traces of every processing step of the traffic.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001202
1203For environments where SNMP is mandatory, at least two agents exist, one is
Davor Ocelic4094ce12017-12-19 23:30:39 +01001204provided with the HAProxy sources and relies on the Net-SNMP Perl module.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001205Another one is provided with the commercial packages and doesn't require Perl.
1206Both are roughly equivalent in terms of coverage.
1207
1208It is often recommended to install 4 utilities on the machine where HAProxy is
1209deployed :
1210
1211 - socat (in order to connect to the CLI, though certain forks of netcat can
1212 also do it to some extents);
1213
1214 - halog from the latest HAProxy version : this is the log analysis tool, it
1215 parses native TCP and HTTP logs extremely fast (1 to 2 GB per second) and
1216 extracts useful information and statistics such as requests per URL, per
1217 source address, URLs sorted by response time or error rate, termination
Davor Ocelic4094ce12017-12-19 23:30:39 +01001218 codes etc. It was designed to be deployed on the production servers to
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001219 help troubleshoot live issues so it has to be there ready to be used;
1220
1221 - tcpdump : this is highly recommended to take the network traces needed to
1222 troubleshoot an issue that was made visible in the logs. There is a moment
1223 where application and haproxy's analysis will diverge and the network traces
1224 are the only way to say who's right and who's wrong. It's also fairly common
1225 to detect bugs in network stacks and hypervisors thanks to tcpdump;
1226
1227 - strace : it is tcpdump's companion. It will report what HAProxy really sees
1228 and will help sort out the issues the operating system is responsible for
1229 from the ones HAProxy is responsible for. Strace is often requested when a
1230 bug in HAProxy is suspected;
1231
1232
12333.4.2. Advanced features : System-specific capabilities
1234-------------------------------------------------------
1235
1236Depending on the operating system HAProxy is deployed on, certain extra features
1237may be available or needed. While it is supported on a number of platforms,
Patrick Starrdce734e2017-10-09 13:17:12 +07001238HAProxy is primarily developed on Linux, which explains why some features are
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001239only available on this platform.
1240
1241The transparent bind and connect features, the support for binding connections
1242to a specific network interface, as well as the ability to bind multiple
1243processes to the same IP address and ports are only available on Linux and BSD
1244systems, though only Linux performs a kernel-side load balancing of the incoming
1245requests between the available processes.
1246
1247On Linux, there are also a number of extra features and optimizations including
1248support for network namespaces (also known as "containers") allowing HAProxy to
1249be a gateway between all containers, the ability to set the MSS, Netfilter marks
1250and IP TOS field on the client side connection, support for TCP FastOpen on the
1251listening side, TCP user timeouts to let the kernel quickly kill connections
1252when it detects the client has disappeared before the configured timeouts, TCP
1253splicing to let the kernel forward data between the two sides of a connections
1254thus avoiding multiple memory copies, the ability to enable the "defer-accept"
1255bind option to only get notified of an incoming connection once data become
1256available in the kernel buffers, and the ability to send the request with the
Davor Ocelic4094ce12017-12-19 23:30:39 +01001257ACK confirming a connect (sometimes called "piggy-back") which is enabled with
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001258the "tcp-smart-connect" option. On Linux, HAProxy also takes great care of
1259manipulating the TCP delayed ACKs to save as many packets as possible on the
1260network.
1261
1262Some systems have an unreliable clock which jumps back and forth in the past
1263and in the future. This used to happen with some NUMA systems where multiple
1264processors didn't see the exact same time of day, and recently it became more
1265common in virtualized environments where the virtual clock has no relation with
1266the real clock, resulting in huge time jumps (sometimes up to 30 seconds have
1267been observed). This causes a lot of trouble with respect to timeout enforcement
1268in general. Due to this flaw of these systems, HAProxy maintains its own
1269monotonic clock which is based on the system's clock but where drift is measured
1270and compensated for. This ensures that even with a very bad system clock, timers
1271remain reasonably accurate and timeouts continue to work. Note that this problem
1272affects all the software running on such systems and is not specific to HAProxy.
1273The common effects are spurious timeouts or application freezes. Thus if this
Davor Ocelic4094ce12017-12-19 23:30:39 +01001274behavior is detected on a system, it must be fixed, regardless of the fact that
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001275HAProxy protects itself against it.
1276
Willy Tarreauec8962c2020-05-05 17:39:16 +02001277On Linux, a new starting process may communicate with the previous one to reuse
1278its listening file descriptors so that the listening sockets are never
Thayne McCombsdab4ba62021-01-07 21:24:41 -07001279interrupted during the process's replacement.
Willy Tarreauec8962c2020-05-05 17:39:16 +02001280
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001281
12823.4.3. Advanced features : Scripting
1283------------------------------------
1284
1285HAProxy can be built with support for the Lua embedded language, which opens a
1286wide area of new possibilities related to complex manipulation of requests or
1287responses, routing decisions, statistics processing and so on. Using Lua it is
1288even possible to establish parallel connections to other servers to exchange
1289information. This way it becomes possible (though complex) to develop an
1290authentication system for example. Please refer to the documentation in the file
1291"doc/lua-api/index.rst" for more information on how to use Lua.
1292
1293
Willy Tarreauec8962c2020-05-05 17:39:16 +020012943.4.4. Advanced features: Tracing
1295---------------------------------
1296
1297At any moment an administrator may connect over the CLI and enable tracing in
1298various internal subsystems. Various levels of details are provided by default
1299so that in practice anything between one line per request to 500 lines per
1300request can be retrieved. Filters as well as an automatic capture on/off/pause
1301mechanism are available so that it really is possible to wait for a certain
1302event and watch it in detail. This is extremely convenient to diagnose protocol
1303violations from faulty servers and clients, or denial of service attacks.
1304
1305
Willy Tarreaud8e42b62015-08-18 21:51:36 +020013063.5. Sizing
1307-----------
1308
1309Typical CPU usage figures show 15% of the processing time spent in HAProxy
1310versus 85% in the kernel in TCP or HTTP close mode, and about 30% for HAProxy
1311versus 70% for the kernel in HTTP keep-alive mode. This means that the operating
1312system and its tuning have a strong impact on the global performance.
1313
1314Usages vary a lot between users, some focus on bandwidth, other ones on request
Davor Ocelic4094ce12017-12-19 23:30:39 +01001315rate, others on connection concurrency, others on SSL performance. This section
1316aims at providing a few elements to help with this task.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001317
1318It is important to keep in mind that every operation comes with a cost, so each
1319individual operation adds its overhead on top of the other ones, which may be
1320negligible in certain circumstances, and which may dominate in other cases.
1321
1322When processing the requests from a connection, we can say that :
1323
1324 - forwarding data costs less than parsing request or response headers;
1325
1326 - parsing request or response headers cost less than establishing then closing
1327 a connection to a server;
1328
1329 - establishing an closing a connection costs less than a TLS resume operation;
1330
1331 - a TLS resume operation costs less than a full TLS handshake with a key
1332 computation;
1333
1334 - an idle connection costs less CPU than a connection whose buffers hold data;
1335
1336 - a TLS context costs even more memory than a connection with data;
1337
1338So in practice, it is cheaper to process payload bytes than header bytes, thus
1339it is easier to achieve high network bandwidth with large objects (few requests
1340per volume unit) than with small objects (many requests per volume unit). This
1341explains why maximum bandwidth is always measured with large objects, while
1342request rate or connection rates are measured with small objects.
1343
Davor Ocelic4094ce12017-12-19 23:30:39 +01001344Some operations scale well on multiple processes spread over multiple CPUs,
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001345and others don't scale as well. Network bandwidth doesn't scale very far because
1346the CPU is rarely the bottleneck for large objects, it's mostly the network
Davor Ocelic4094ce12017-12-19 23:30:39 +01001347bandwidth and data buses to reach the network interfaces. The connection rate
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001348doesn't scale well over multiple processors due to a few locks in the system
1349when dealing with the local ports table. The request rate over persistent
1350connections scales very well as it doesn't involve much memory nor network
1351bandwidth and doesn't require to access locked structures. TLS key computation
1352scales very well as it's totally CPU-bound. TLS resume scales moderately well,
1353but reaches its limits around 4 processes where the overhead of accessing the
1354shared table offsets the small gains expected from more power.
1355
1356The performance numbers one can expect from a very well tuned system are in the
1357following range. It is important to take them as orders of magnitude and to
1358expect significant variations in any direction based on the processor, IRQ
1359setting, memory type, network interface type, operating system tuning and so on.
1360
Davor Ocelic4094ce12017-12-19 23:30:39 +01001361The following numbers were found on a Core i7 running at 3.7 GHz equipped with
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001362a dual-port 10 Gbps NICs running Linux kernel 3.10, HAProxy 1.6 and OpenSSL
13631.0.2. HAProxy was running as a single process on a single dedicated CPU core,
1364and two extra cores were dedicated to network interrupts :
1365
1366 - 20 Gbps of maximum network bandwidth in clear text for objects 256 kB or
1367 higher, 10 Gbps for 41kB or higher;
1368
1369 - 4.6 Gbps of TLS traffic using AES256-GCM cipher with large objects;
1370
1371 - 83000 TCP connections per second from client to server;
1372
1373 - 82000 HTTP connections per second from client to server;
1374
1375 - 97000 HTTP requests per second in server-close mode (keep-alive with the
1376 client, close with the server);
1377
1378 - 243000 HTTP requests per second in end-to-end keep-alive mode;
1379
1380 - 300000 filtered TCP connections per second (anti-DDoS)
1381
1382 - 160000 HTTPS requests per second in keep-alive mode over persistent TLS
1383 connections;
1384
1385 - 13100 HTTPS requests per second using TLS resumed connections;
1386
Davor Ocelic4094ce12017-12-19 23:30:39 +01001387 - 1300 HTTPS connections per second using TLS connections renegotiated with
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001388 RSA2048;
1389
1390 - 20000 concurrent saturated connections per GB of RAM, including the memory
1391 required for system buffers; it is possible to do better with careful tuning
Davor Ocelic4094ce12017-12-19 23:30:39 +01001392 but this result it easy to achieve.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001393
1394 - about 8000 concurrent TLS connections (client-side only) per GB of RAM,
1395 including the memory required for system buffers;
1396
1397 - about 5000 concurrent end-to-end TLS connections (both sides) per GB of
1398 RAM including the memory required for system buffers;
1399
1400Thus a good rule of thumb to keep in mind is that the request rate is divided
1401by 10 between TLS keep-alive and TLS resume, and between TLS resume and TLS
Davor Ocelic4094ce12017-12-19 23:30:39 +01001402renegotiation, while it's only divided by 3 between HTTP keep-alive and HTTP
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001403close. Another good rule of thumb is to remember that a high frequency core
1404with AES instructions can do around 5 Gbps of AES-GCM per core.
1405
Patrick Starrdce734e2017-10-09 13:17:12 +07001406Having more cores rarely helps (except for TLS) and is even counter-productive
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001407due to the lower frequency. In general a small number of high frequency cores
1408is better.
1409
1410Another good rule of thumb is to consider that on the same server, HAProxy will
1411be able to saturate :
1412
1413 - about 5-10 static file servers or caching proxies;
1414
1415 - about 100 anti-virus proxies;
1416
Willy Tarreau16af23c2015-08-27 16:30:53 +02001417 - and about 100-1000 application servers depending on the technology in use.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001418
1419
14203.6. How to get HAProxy
1421-----------------------
1422
Davor Ocelic4094ce12017-12-19 23:30:39 +01001423HAProxy is an open source project covered by the GPLv2 license, meaning that
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001424everyone is allowed to redistribute it provided that access to the sources is
1425also provided upon request, especially if any modifications were made.
1426
1427HAProxy evolves as a main development branch called "master" or "mainline", from
1428which new branches are derived once the code is considered stable. A lot of web
1429sites run some development branches in production on a voluntarily basis, either
1430to participate to the project or because they need a bleeding edge feature, and
1431their feedback is highly valuable to fix bugs and judge the overall quality and
Patrick Starrdce734e2017-10-09 13:17:12 +07001432stability of the version being developed.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001433
1434The new branches that are created when the code is stable enough constitute a
1435stable version and are generally maintained for several years, so that there is
1436no emergency to migrate to a newer branch even when you're not on the latest.
1437Once a stable branch is issued, it may only receive bug fixes, and very rarely
1438minor feature updates when that makes users' life easier. All fixes that go into
1439a stable branch necessarily come from the master branch. This guarantees that no
1440fix will be lost after an upgrade. For this reason, if you fix a bug, please
1441make the patch against the master branch, not the stable branch. You may even
1442discover it was already fixed. This process also ensures that regressions in a
1443stable branch are extremely rare, so there is never any excuse for not upgrading
1444to the latest version in your current branch.
1445
Willy Tarreauec8962c2020-05-05 17:39:16 +02001446Branches are numbered with two digits delimited with a dot, such as "1.6".
1447Since 1.9, branches with an odd second digit are mostly focused on sensitive
1448technical updates and more aimed at advanced users because they are likely to
1449trigger more bugs than the other ones. They are maintained for about a year
1450only and must not be deployed where they cannot be rolled back in emergency. A
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001451complete version includes one or two sub-version numbers indicating the level of
1452fix. For example, version 1.5.14 is the 14th fix release in branch 1.5 after
1453version 1.5.0 was issued. It contains 126 fixes for individual bugs, 24 updates
1454on the documentation, and 75 other backported patches, most of which were needed
Patrick Starrdce734e2017-10-09 13:17:12 +07001455to fix the aforementioned 126 bugs. An existing feature may never be modified
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001456nor removed in a stable branch, in order to guarantee that upgrades within the
1457same branch will always be harmless.
1458
1459HAProxy is available from multiple sources, at different release rhythms :
1460
1461 - The official community web site : http://www.haproxy.org/ : this site
1462 provides the sources of the latest development release, all stable releases,
1463 as well as nightly snapshots for each branch. The release cycle is not fast,
1464 several months between stable releases, or between development snapshots.
1465 Very old versions are still supported there. Everything is provided as
1466 sources only, so whatever comes from there needs to be rebuilt and/or
1467 repackaged;
1468
Willy Tarreauec8962c2020-05-05 17:39:16 +02001469 - GitHub : https://github.com/haproxy/haproxy/ : this is the mirror for the
1470 development branch only, which provides integration with the issue tracker,
1471 continuous integration and code coverage tools. This is exclusively for
1472 contributors;
1473
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001474 - A number of operating systems such as Linux distributions and BSD ports.
1475 These systems generally provide long-term maintained versions which do not
1476 always contain all the fixes from the official ones, but which at least
1477 contain the critical fixes. It often is a good option for most users who do
1478 not seek advanced configurations and just want to keep updates easy;
1479
1480 - Commercial versions from http://www.haproxy.com/ : these are supported
1481 professional packages built for various operating systems or provided as
1482 appliances, based on the latest stable versions and including a number of
1483 features backported from the next release for which there is a strong
1484 demand. It is the best option for users seeking the latest features with
1485 the reliability of a stable branch, the fastest response time to fix bugs,
Davor Ocelic4094ce12017-12-19 23:30:39 +01001486 or simply support contracts on top of an open source product;
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001487
1488
1489In order to ensure that the version you're using is the latest one in your
1490branch, you need to proceed this way :
1491
1492 - verify which HAProxy executable you're running : some systems ship it by
1493 default and administrators install their versions somewhere else on the
1494 system, so it is important to verify in the startup scripts which one is
1495 used;
1496
1497 - determine which source your HAProxy version comes from. For this, it's
1498 generally sufficient to type "haproxy -v". A development version will
1499 appear like this, with the "dev" word after the branch number :
1500
1501 HA-Proxy version 1.6-dev3-385ecc-68 2015/08/18
1502
1503 A stable version will appear like this, as well as unmodified stable
1504 versions provided by operating system vendors :
1505
1506 HA-Proxy version 1.5.14 2015/07/02
1507
1508 And a nightly snapshot of a stable version will appear like this with an
1509 hexadecimal sequence after the version, and with the date of the snapshot
1510 instead of the date of the release :
1511
1512 HA-Proxy version 1.5.14-e4766ba 2015/07/29
1513
1514 Any other format may indicate a system-specific package with its own
1515 patch set. For example HAProxy Enterprise versions will appear with the
1516 following format (<branch>-<latest commit>-<revision>) :
1517
1518 HA-Proxy version 1.5.0-994126-357 2015/07/02
1519
Willy Tarreauec8962c2020-05-05 17:39:16 +02001520 In addition, versions 2.1 and above will include a "Status" line indicating
1521 whether the version is safe for production or not, and if so, till when, as
1522 well as a link to the list of known bugs affecting this version.
1523
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001524 - for system-specific packages, you have to check with your vendor's package
1525 repository or update system to ensure that your system is still supported,
1526 and that fixes are still provided for your branch. For community versions
1527 coming from haproxy.org, just visit the site, verify the status of your
1528 branch and compare the latest version with yours to see if you're on the
1529 latest one. If not you can upgrade. If your branch is not maintained
1530 anymore, you're definitely very late and will have to consider an upgrade
1531 to a more recent branch (carefully read the README when doing so).
1532
1533HAProxy will have to be updated according to the source it came from. Usually it
1534follows the system vendor's way of upgrading a package. If it was taken from
1535sources, please read the README file in the sources directory after extracting
1536the sources and follow the instructions for your operating system.
1537
1538
15394. Companion products and alternatives
1540--------------------------------------
1541
1542HAProxy integrates fairly well with certain products listed below, which is why
Davor Ocelic4094ce12017-12-19 23:30:39 +01001543they are mentioned here even if not directly related to HAProxy.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001544
1545
15464.1. Apache HTTP server
1547-----------------------
1548
1549Apache is the de-facto standard HTTP server. It's a very complete and modular
1550project supporting both file serving and dynamic contents. It can serve as a
Michael Prokop4438c602019-05-24 10:25:45 +02001551frontend for some application servers. It can even proxy requests and cache
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001552responses. In all of these use cases, a front load balancer is commonly needed.
Patrick Starrdce734e2017-10-09 13:17:12 +07001553Apache can work in various modes, some being heavier than others. Certain
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001554modules still require the heavier pre-forked model and will prevent Apache from
1555scaling well with a high number of connections. In this case HAProxy can provide
1556a tremendous help by enforcing the per-server connection limits to a safe value
1557and will significantly speed up the server and preserve its resources that will
1558be better used by the application.
1559
1560Apache can extract the client's address from the X-Forwarded-For header by using
1561the "mod_rpaf" extension. HAProxy will automatically feed this header when
1562"option forwardfor" is specified in its configuration. HAProxy may also offer a
1563nice protection to Apache when exposed to the internet, where it will better
Davor Ocelic4094ce12017-12-19 23:30:39 +01001564resist a wide number of types of DoS attacks.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001565
1566
15674.2. NGINX
1568----------
1569
1570NGINX is the second de-facto standard HTTP server. Just like Apache, it covers a
1571wide range of features. NGINX is built on a similar model as HAProxy so it has
1572no problem dealing with tens of thousands of concurrent connections. When used
Davor Ocelic4094ce12017-12-19 23:30:39 +01001573as a gateway to some applications (e.g. using the included PHP FPM) it can often
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001574be beneficial to set up some frontend connection limiting to reduce the load
1575on the PHP application. HAProxy will clearly be useful there both as a regular
Davor Ocelic4094ce12017-12-19 23:30:39 +01001576load balancer and as the traffic regulator to speed up PHP by decongesting
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001577it. Also since both products use very little CPU thanks to their event-driven
1578architecture, it's often easy to install both of them on the same system. NGINX
1579implements HAProxy's PROXY protocol, thus it is easy for HAProxy to pass the
1580client's connection information to NGINX so that the application gets all the
1581relevant information. Some benchmarks have also shown that for large static
1582file serving, implementing consistent hash on HAProxy in front of NGINX can be
1583beneficial by optimizing the OS' cache hit ratio, which is basically multiplied
1584by the number of server nodes.
1585
1586
15874.3. Varnish
1588------------
1589
1590Varnish is a smart caching reverse-proxy, probably best described as a web
1591application accelerator. Varnish doesn't implement SSL/TLS and wants to dedicate
1592all of its CPU cycles to what it does best. Varnish also implements HAProxy's
1593PROXY protocol so that HAProxy can very easily be deployed in front of Varnish
1594as an SSL offloader as well as a load balancer and pass it all relevant client
1595information. Also, Varnish naturally supports decompression from the cache when
1596a server has provided a compressed object, but doesn't compress however. HAProxy
1597can then be used to compress outgoing data when backend servers do not implement
1598compression, though it's rarely a good idea to compress on the load balancer
1599unless the traffic is low.
1600
1601When building large caching farms across multiple nodes, HAProxy can make use of
1602consistent URL hashing to intelligently distribute the load to the caching nodes
1603and avoid cache duplication, resulting in a total cache size which is the sum of
Willy Tarreauec8962c2020-05-05 17:39:16 +02001604all caching nodes. In addition, caching of very small dumb objects for a short
1605duration on HAProxy can sometimes save network round trips and reduce the CPU
1606load on both the HAProxy and the Varnish nodes. This is only possible is no
1607processing is done on these objects on Varnish (this is often referred to as
1608the notion of "favicon cache", by which a sizeable percentage of useless
1609downstream requests can sometimes be avoided). However do not enable HAProxy
1610caching for a long time (more than a few seconds) in front of any other cache,
1611that would significantly complicate troubleshooting without providing really
1612significant savings.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001613
1614
16154.4. Alternatives
1616-----------------
1617
1618Linux Virtual Server (LVS or IPVS) is the layer 4 load balancer included within
1619the Linux kernel. It works at the packet level and handles TCP and UDP. In most
1620cases it's more a complement than an alternative since it doesn't have layer 7
1621knowledge at all.
1622
1623Pound is another well-known load balancer. It's much simpler and has much less
1624features than HAProxy but for many very basic setups both can be used. Its
1625author has always focused on code auditability first and wants to maintain the
1626set of features low. Its thread-based architecture scales less well with high
1627connection counts, but it's a good product.
1628
1629Pen is a quite light load balancer. It supports SSL, maintains persistence using
1630a fixed-size table of its clients' IP addresses. It supports a packet-oriented
1631mode allowing it to support direct server return and UDP to some extents. It is
1632meant for small loads (the persistence table only has 2048 entries).
1633
1634NGINX can do some load balancing to some extents, though it's clearly not its
1635primary function. Production traffic is used to detect server failures, the
1636load balancing algorithms are more limited, and the stickiness is very limited.
1637But it can make sense in some simple deployment scenarios where it is already
1638present. The good thing is that since it integrates very well with HAProxy,
Davor Ocelic4094ce12017-12-19 23:30:39 +01001639there's nothing wrong with adding HAProxy later when its limits have been
1640reached.
Willy Tarreaud8e42b62015-08-18 21:51:36 +02001641
1642Varnish also does some load balancing of its backend servers and does support
1643real health checks. It doesn't implement stickiness however, so just like with
1644NGINX, as long as stickiness is not needed that can be enough to start with.
1645And similarly, since HAProxy and Varnish integrate so well together, it's easy
1646to add it later into the mix to complement the feature set.
1647
Willy Tarreau65626232020-05-05 18:08:07 +02001648
16495. Contacts
1650-----------
1651
1652If you want to contact the developers or any community member about anything,
1653the best way to do it usually is via the mailing list by sending your message
1654to haproxy@formilux.org. Please note that this list is public and its archives
1655are public as well so you should avoid disclosing sensitive information. A
1656thousand of users of various experience levels are present there and even the
1657most complex questions usually find an optimal response relatively quickly.
1658Suggestions are welcome too. For users having difficulties with e-mail, a
1659Discourse platform is available at http://discourse.haproxy.org/ . However
1660please keep in mind that there are less people reading questions there and that
1661most are handled by a really tiny team. In any case, please be patient and
1662respectful with those who devote their spare time helping others.
1663
1664I you believe you've found a bug but are not sure, it's best reported on the
1665mailing list. If you're quite convinced you've found a bug, that your version
1666is up-to-date in its branch, and you already have a GitHub account, feel free
1667to go directly to https://github.com/haproxy/haproxy/ and file an issue with
1668all possibly available details. Again, this is public so be careful not to post
1669information you might later regret. Since the issue tracker presents itself as
1670a very long thread, please avoid pasting very long dumps (a few hundreds lines
1671or more) and attach them instead.
1672
1673If you've found what you're absolutely certain can be considered a critical
1674security issue that would put many users in serious trouble if discussed in a
1675public place, then you can send it with the reproducer to security@haproxy.org.
1676A small team of trusted developers will receive it and will be able to propose
1677a fix. We usually don't use embargoes and once a fix is available it gets
1678merged. In some rare circumstances it can happen that a release is coordinated
1679with software vendors. Please note that this process usually messes up with
1680eveyone's work, and that rushed up releases can sometimes introduce new bugs,
1681so it's best avoided unless strictly necessary; as such, there is often little
1682consideration for reports that needlessly cause such extra burden, and the best
1683way to see your work credited usually is to provide a working fix, which will
1684appear in changelogs.