NAME Perlbal::Plugin::SessionAffinity - Sane session affinity (sticky sessions) for Perlbal VERSION version 0.100 SYNOPSIS LOAD SessionAffinity CREATE POOL backends POOL backends ADD 10.20.20.100 POOL backends ADD 10.20.20.101 POOL backends ADD 10.20.20.102 CREATE SERVICE balancer SET listen = 0.0.0.0:80 SET role = reverse_proxy SET pool = backends SET persist_client = on SET persist_backend = on SET verify_backend = on SET plugins = sessionaffinity ENABLE balancer DESCRIPTION Perlbal doesn't support session affinity (or otherwise known as "sticky sessions") out of the box. There is a plugin on CPAN called Perlbal::Plugin::StickySessions but there are a few problems with it. This plugin should be do a much better job. Go ahead and read why you should use this one and how it works. WHY YOU SHOULD USE IT Here are things that are unique in this plugin. I am comparing this with the current available session affinity implementation available on CPAN (Perlbal::Plugin::StickySessions). * It supports session affinity for all requests Unlike the other plugin, this one uses a proper hook that supports not just file fetching, but for each and every request. * No patches required Unlike the other plugin, that comes with two patches (which were not integrated into Perlbal core), this one requires no patches whatsoever. * It's up-to-date Unlike the other plugin, that still requires a patch that includes a hook that was already introduced (which shows it's clearly outdated), this plugin is very much up to speed with things. * It's thin and sane Unlike the other plugin, which is mostly copy-pasted from some handling code in Perlbal itself (seriously!), this module contains no copy-pasted code, is much smaller and leaner, and is much less likely to break between new versions of Perlbal. * No breakage Unlike the other plugin, which - after close inspection - seemed breakable (to say the least, since connect-aheads don't seem to get cleaned up), this plugin uses a completely different method which emphasizes correctness and the least intervention with Perlbal itself, and keeps Perlbal in charge of the critical operations. Small note here: this does not mean it will definitely play nice with everything you already have. Specifically any hooks that rely on the name of the service might be affected. Please read further under Incompatibilities to understand the issue better. * Much less security risk Unlike the other plugin, which sets a cookie with the backend ID correlating to the backend order in the pool, this plugin uses SHA1 checksum IDs (with an optionally randomly-created salt) for each server, and allows you to change the header name and add a checksum salt (whether randomly-created or your own) for the cookie. This makes it harder for an attacker to understand what the header represents and how many backends exist (since there is no counter). * Features Unlike the other plugin, that simply has things hardcoded, this plugin allows to change both the header name and the salt used to create the ID. By default the salt is off but you can turn it on and then either use a randomly-created one or set your own. HOW DOES IT WORK Basic stuff Basically, the module creates a SHA1 checksum for each backend node, and provides the user with a cookie request. If the user provides that cookie in return, it will try and find and provide the user with that specific node. If the node is no longer in the service's pool, or the cookie matches a node that doesn't exist, it will provide the user with a cookie again. Advanced stuff The plugin sets up dedicated pools and services for each service's node. This is required since Perlbal has no way of actually allowing you to specify the node a user will go to, only the service. Not to worry, this creation is done lazily so it saves as much memory as it can. When a user comes in with a cookie of a node that exist in the service's pool it will create a pool for it (if one doesn't exist), and a matching service for it (if one doesn't exist) and then direct to user to it. The check against nodes and pools is done live and not against the static configuration file. This means that if you're playing with the pools (changing them live, for example), it will still work just fine. A new service is created using configurations from the existing service. The more interesting details is that reuse is emphasized so no new sockets are created and instead this new service uses the already existing sockets (along with existing connections) instead of firing new ones. It doesn't open a new socket for listening or anything like that. This also means your SSL connections work seamlessly. Yes, it's insanely cool, I know! :) Incompatibilities If you've read the Advanced stuff section above, you might have guessed a possible problem with anything that relies on the name of the service. If you're using a plugin that relies on the name of the service, you might notice it stops working properly. This is because the new service that is generated by SessionAffinity is no longer the previous service, and doesn't contain its name. Instead it has its own name, which is not known to your plugin. If you're using the "header" command to add headers to the backend, fear not. We copy over the headers from the original service to the new one. That still works just fine. One possible way to fix it (implemented and later removed) is to include the previous name in a new unofficial (and unauthorized) key in the service hash. ATTRIBUTES affinity_cookie_header The name of the cookie header for the session. Default: X-SERVERID. affinity_use_salt Whether to use a salt or not when calculating SHA1 IDs. # both are equal affinity_use_salt = 1 affinity_use_salt = yes # opposite meaning affinity_use_salt = 0 affinity_use_salt = no Default: no. affinity_salt The salt that is used to create the backend's SHA1 IDs. Default: the following code is run when you load Perlbal::Plugin::SessionAffinity to create the salt on start up: join q{}, map { $_ = rand 999; s/\.//; $_ } 1 .. 10; If you want predictability with salt, you can override it as such: affinity_salt = helloworld # now the calculation will be: my $sha1 = sha1hex( $salt . $ip . $port ); affinity_use_domain Uses domain-mode for finding the backend. This is an alternate way of deciding the backend, which enables backends to persist per domain, allowing you to avoid a fragmented cache. If you have a lot of cache misses because of jumping between backends, try turning this feature on. This feature ignores the cookie provided (and does not provide its own cookie) since backends are decided by the domain name alone. # both are equal affinity_use_domain = 1 affinity_use_domain = yes # opposite meaning affinity_use_domain = 0 affinity_use_domain = no Default: no. affinity_use_dynamic_cookie Allows you to provide domain-specific dynamic cookies. In case you're serving multiple domains and you want each one to have its own cookie, you can turn this feature on, and it will generate a cookie that is an MD5 of the domain, then wrapped in base64. # both are equal affinity_use_dynamic_cookie = 1 affinity_use_dynamic_cookie = yes # opposite meaning affinity_use_dynamic_cookie = 0 affinity_use_dynamic_cookie = no Default: no. SUBROUTINES/METHODS register Registers our events. unregister Unregister our hooks and setters events. get_ip_port Parses a request's cookies and finds the specific cookie relating to session affinity and get the backend details via the ID in the cookie. find_backend_by_id Given a SHA1 ID, find the correct backend to which it belongs. find_backend_by_domain_id Given a SHA1 ID for a domain, find the correct backend to which it belongs. create_id Creates a SHA1 checksum ID using Digest::SHA. The checksum is composed of the IP, port and salt. If you want to have more predictability, you can provide a salt of 0 or "string" and then the checksum would be predictable. This should make it clear on how it's created: if ( $has_salt ) { $checksum = sha1sum( $salt . "$ip:$port" ); } else { $checksum = sha1sum( "$ip:$port" ); } create_domain_id Same concept as the above "create_id" function, except for the following changes: Accepts a domain and a list of nodes (which is assumed to be ordered), uses the "domain_index" function to get the index in the nodes of a domain and picks the correct node from the list it receives by index. domain_index This function tries to fetch an index number for a given domain name. It accepts a domain name and the maximum index number. It translates the domain name to a long number, and uses mod ("%") on it. DEPENDENCIES Perlbal Obviously. CGI::Cookies To parse and create cookies. Digest::SHA To provide a SHA1 checksum. SEE ALSO Perlbal::Plugin::StickySessions AUTHOR Sawyer X COPYRIGHT AND LICENSE This software is copyright (c) 2014 by Sawyer X. This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.