Redis cache backends ==================== This package provides two different Redis cache backends. If you want to use Redis as cache backend, you have to choose one of the two, but you cannot use both at the same time. Well, it will be technically possible, but it would be quite a dumb thing to do. Predis ------ This implementation uses the Predis PHP library. It is compatible PHP 5.3 only. PhpRedis -------- This implementation uses the PhpRedis PHP extention. In order to use it, you probably will need to compile the extension yourself. Redis version ------------- This module requires Redis version to be 2.6.0 or later with LUA scrpting enabled due to the EVAL command usage. If you can't upgrade you Redis server: - 3.x release will only officially support Redis server <= 2.8 and 3.x nevertheless you may use it with Redis 2.4 if you configure your cache backend to operate in sharding mode. - For Redis 2.4 use the latest 2.x release of this module or use the 3.x release with sharding mode enabled. - For Redis <=2.3 use any version of this module <=2.6 Notes ----- Both backends provide the exact same functionalities. The major difference is because PhpRedis uses a PHP extension, and not PHP code, it will performe a lot better (Predis needs PHP userland code to be loaded). Difference is not that visible, it's really a few millisec on my testing box, in case you attempt to profile the code, traces will be a lot bigger. Note that most of the settings are shared. See next sections. Getting started =============== Quick setup ----------- Here is a simple yet working easy way to setup the module. This method will allow Drupal to use Redis for all caches and locks and path alias cache replacement. $conf['redis_client_interface'] = 'PhpRedis'; // Can be "Predis". $conf['redis_client_host'] = '1.2.3.4'; // Your Redis instance hostname. $conf['lock_inc'] = 'sites/all/modules/redis/redis.lock.inc'; $conf['path_inc'] = 'sites/all/modules/redis/redis.path.inc'; $conf['cache_backends'][] = 'sites/all/modules/redis/redis.autoload.inc'; $conf['cache_default_class'] = 'Redis_Cache'; See next chapters for more information. Is there any cache bins that should *never* go into Redis? ---------------------------------------------------------- TL;DR: No. Except for 'cache_form' if you use Redis with LRU eviction. Redis has been maturing a lot over time, and will apply different sensible settings for different bins; It's today very stable. Advanced configuration ====================== Use the compressed cache ------------------------ Please note this is for now an experimental feature. As a personnal note from the module author, it should be safe to use. Use this cache class setting to enable compression. This will save usually about 80% RAM at the cost of some milliseconds server time. $conf['cache_default_class'] = 'Redis_CacheCompressed'; Additionnaly, you can alter the default size compression threshold, under which entries will not be compressed (size is in bytes, set 0 to always compress): $conf['cache_compression_size_threshold'] = 100; You can also change the compression level, which an positive integer between 1 and 9, 1 being the lowest but fastest compression ratio, 9 being the most aggressive compression but is a lot slower. From testing, setting it to the lower level (1) gives 80% memory usage decrease, which is more than enough. $conf['cache_compression_ratio'] = 5; Please note that those settings are global and not on a cache bin basis, you can already control whenever the compression is to be used or not by selecting a different cache class on per cache bin basis. If you switch from the standard default backend (without compression) to the compressed cache backend, it will recover transparently uncompressed data and proceed normally without additional cache eviction, it safe to upgrade. Donwgrading from compressed data to uncompressed data won't work, but the cache backend will just give you cache hit miss and it will work seamlessly too without any danger for the site. Choose the Redis client library to use -------------------------------------- Add into your settings.php file: $conf['redis_client_interface'] = 'PhpRedis'; You can replace 'PhpRedis' with 'Predis', depending on the library you chose. Note that this is optional but recommended. If you don't set this variable the module will proceed to class lookups and attempt to choose the best client available, having always a preference for the Predis one. Tell Drupal to use the cache backend ------------------------------------ Usual cache backend configuration, as follows, to add into your settings.php file like any other backend: $conf['cache_backends'][] = 'sites/all/modules/redis/redis.autoload.inc'; $conf['cache_class_cache'] = 'Redis_Cache'; $conf['cache_class_cache_menu'] = 'Redis_Cache'; $conf['cache_class_cache_bootstrap'] = 'Redis_Cache'; // ... Any other bins. Tell Drupal to use the lock backend ----------------------------------- Usual lock backend override, update you settings.php file as this: $conf['lock_inc'] = 'sites/all/modules/redis/redis.lock.inc'; Tell Drupal to use the path alias backend ----------------------------------------- Usual path backend override, update you settings.php file as this: $conf['path_inc'] = 'sites/all/modules/redis/redis.path.inc'; Notice that there is an additional variable for path handling that is set per default which will ignore any path that is an admin path, gaining a few SQL queries. If you want to be able to set aliases on admin path and restore an almost default Drupal core behavior, you should add this line into your settings.php file: $conf['path_alias_admin_blacklist'] = FALSE; Drupal 6 and lock backend ------------------------- Considering this is a Drupal 7 module only downloading it in Drupal 6 will make the module UI telling you this module is unsupported yet you can use the lock backend on Drupal 6. Read your Drupal 6 core documentation and use the redis.lock.inc file as lock_inc replacement the same way its being done for Drupal 7 and it should work. Note that this is untested by the module maintainer (feedback will be greatly appreciated). Common settings =============== Connect throught a UNIX socket ------------------------------ All you have to do is specify this line: $conf['redis_client_socket'] = '/some/path/redis.sock'; Both drivers support it. Connect to a remote host ------------------------ If your Redis instance is remote, you can use this syntax: $conf['redis_client_host'] = '1.2.3.4'; $conf['redis_client_port'] = 1234; Port is optional, default is 6379 (default Redis port). Using a specific database ------------------------- Per default, Redis ships the database "0". All default connections will be use this one if nothing is specified. Depending on you OS or OS distribution, you might have numerous database. To use one in particular, just add to your settings.php file: $conf['redis_client_base'] = 12; Please note that if you are working in shard mode, you should never set this variable. Connection to a password protected instance ------------------------------------------- If you are using a password protected instance, specify the password this way: $conf['redis_client_password'] = "mypassword"; Depending on the backend, using a wrong auth will behave differently: - Predis will throw an exception and make Drupal fail during early boostrap. - PhpRedis will make Redis calls silent and creates some PHP warnings, thus Drupal will behave as if it was running with a null cache backend (no cache at all). Prefixing site cache entries (avoiding sites name collision) ------------------------------------------------------------ If you need to differenciate multiple sites using the same Redis instance and database, you will need to specify a prefix for your site cache entries. Important note: most people don't need that feature since that when no prefix is specified, the Redis module will attempt to use the a hash of the database credentials in order to provide a multisite safe default behavior. This means that the module will also safely work in CLI scripts. Cache prefix configuration attemps to use a unified variable accross contrib backends that support this feature. This variable name is 'cache_prefix'. This variable is polymorphic, the simplest version is to provide a raw string that will be the default prefix for all cache bins: $conf['cache_prefix'] = 'mysite_'; Alternatively, to provide the same functionality, you can provide the variable as an array: $conf['cache_prefix']['default'] = 'mysite_'; This allows you to provide different prefix depending on the bin name. Common usage is that each key inside the 'cache_prefix' array is a bin name, the value the associated prefix. If the value is explicitely FALSE, then no prefix is used for this bin. The 'default' meta bin name is provided to define the default prefix for non specified bins. It behaves like the other names, which means that an explicit FALSE will order the backend not to provide any prefix for any non specified bin. Here is a complex sample: // Default behavior for all bins, prefix is 'mysite_'. $conf['cache_prefix']['default'] = 'mysite_'; // Set no prefix explicitely for 'cache' and 'cache_bootstrap' bins. $conf['cache_prefix']['cache'] = FALSE; $conf['cache_prefix']['cache_bootstrap'] = FALSE; // Set another prefix for 'cache_menu' bin. $conf['cache_prefix']['cache_menu'] = 'menumysite_'; Note that this last notice is Redis only specific, because per default Redis server will not namespace data, thus sharing an instance for multiple sites will create conflicts. This is not true for every contributed backends. Sharding vs normal mode ----------------------- Per default the Redis cache backend will be in "normal" mode, meaning that every flush call will trigger and EVAL lua script that will proceed to cache wipeout and cleanup the Redis database from stalled entries. Nevertheless, if you are working with a Redis server < 2.6 or in a sharded environment, you cannot multiple keys per command nor proceed to EVAL'ed scripts, you will then need to switch to the sharded mode. Sharded mode will never delete entries on flush calls, but register a key with the current flush time instead. Cache entries will then be deleted on read if the entry checksum does not match or is older than the latest flush call. Note that this mode is fast and safe, but must be used accordingly with the default lifetime for permanent items, else your Redis server might keep stalled entries into its database forever. In order to enable the sharded mode, set into your settings.php file: $conf['redis_flush_mode'] = 3; Please note that the value 3 is there to keep backward compatibility with older versions of the Redis module and will not change. Note that previous Redis module version allowed to set a per-bin setting for the clear mode value; nevertheless the clear mode is not a valid setting anymore and the past issues have been resolved. Only the global value will work as of now. Sharding and pipelining ----------------------- Whe using this module with sharding mode you may have a sharding proxy able to do command pipelining. If that is the case, you should switch to "sharding with pipelining" mode instead: $conf['redis_flush_mode'] = 4; Note that if you use the sharding mode because you use an older version of the Redis server, you should always use this mode to ensure the best performances. Default lifetime for permanent items ------------------------------------ Redis when reaching its maximum memory limit will stop writing data in its storage engine: this is a feature that avoid the Redis server crashing when there is no memory left on the machine. As a workaround, Redis can be configured as a LRU cache for both volatile or permanent items, which means it can behave like Memcache; Problem is that if you use Redis as a permanent storage for other business matters than this module you cannot possibly configure it to drop permanent items or you'll loose data. This workaround allows you to explicity set a very long or configured default lifetime for CACHE_PERMANENT items (that would normally be permanent) which will mark them as being volatile in Redis storage engine: this then allows you to configure a LRU behavior for volatile keys without engaging the permenent business stuff in a dangerous LRU mechanism; Cache items even if permament will be dropped when unused using this. Per default the TTL for permanent items will set to safe-enough value which is one year; No matter how Redis will be configured default configuration or lazy admin will inherit from a safe module behavior with zero-conf. For advanturous people, you can manage the TTL on a per bin basis and change the default one: // Make CACHE_PERMANENT items being permanent once again // 0 is a special value usable for all bins to explicitely tell the // cache items will not be volatile in Redis. $conf['redis_perm_ttl'] = 0; // Make them being volatile with a default lifetime of 1 year. $conf['redis_perm_ttl'] = "1 year"; // You can override on a per-bin basis; // For example make cached field values live only 3 monthes: $conf['redis_perm_ttl_cache_field'] = "3 months"; // But you can also put a timestamp in there; In this case the // value must be a STRICTLY TYPED integer: $conf['redis_perm_ttl_cache_field'] = 2592000; // 30 days. Time interval string will be parsed using DateInterval::createFromDateString please refer to its documentation: http://www.php.net/manual/en/dateinterval.createfromdatestring.php Please also be careful about the fact that those settings are overriden by the 'cache_lifetime' Drupal variable, which should always be set to 0. Moreover, this setting will affect all cache entries without exception so be careful and never set values too low if you don't want this setting to override default expire value given by modules on temporary cache entries. Lock backends ------------- Both implementations provides a Redis lock backend. Redis lock backend proved to be faster than the default SQL based one when using both servers on the same box. Both backends, thanks to the Redis WATCH, MULTI and EXEC commands provides a real race condition free mutexes by using Redis transactions. Queue backend ------------- This module provides an experimental queue backend. It is for now implemented only using the PhpRedis driver, any attempt to use it using Predis will result in runtime errors. If you want to change the queue driver system wide, set this into your setting.php file: $conf['queue_default_class'] = 'Redis_Queue'; $conf['queue_default_reliable_class'] = 'Redis_Queue'; Note that some queue implementations such as the batch queue are hardcoded within Drupal and will always use a database dependent implementation. If you need to proceed with finer tuning, you can set a per-queue class in such way: $conf['queue_class_NAME'] = 'Redis_Queue'; Where NAME is the arbitrary module given queue name, used as first parameter for the method DrupalQueue::get(). THIS IS STILL VERY EXPERIMENTAL. The queue should work without any problems except it does not implement the item lease time correctly, this means that items that are too long to process won't be released back and forth but will block the thread processing it instead. This is the only side effect I am aware of at the current time. Failover, sharding and partionning ================================== Important notice ---------------- There are numerous support and feature request issues about client sharding, failover ability, multi-server connection, ability to read from slave and server clustering opened in the issue queue. Note that there is not one universally efficient solution for this: most of the solutions require that you cannot use the MULTI/EXEC command using more than one key, and that you cannot use complex UNION and intersection features anymore. This module does not implement any kind of client side key hashing or sharding and never intended to; We recommend that you read the official Redis documentation page about partionning. The best solution for clustering and sharding today seems to be the proxy assisted partionning using tools such as Twemproxy. Current components state ------------------------ As of now, provided components are simple enough so they never use WATCH or MULTI/EXEC transaction blocks on multiple keys : this means that you can use them in an environment doing data sharding/partionning. This remains true except when you use a proxy that blocks those commands such as Twemproxy. Lock ---- Lock backend works on a single key per lock, it theorically guarantees the atomicity of operations therefore is usable in a sharded environement. Sadly if you use proxy assisted sharding such as Twemproxy, WATCH, MULTI and EXEC commands won't pass making it non shardable. Path ---- Path backend does not use on transactions, it is safe to use in a sharded environment. Note that this backend uses a single HASH key per language and per way (alias to source or source to alias) and therefore won't benefit greatly if not at all from being sharded. Cache ----- Cache uses pipelined transactions but does not uses it to guarantee any kind of data consistency. If you use a smart sharding proxy it is supposed to work transparently without any hickups. Queue ----- Queue is still in development. There might be problems in the long term for this component in sharded environments. Testing ======= Due to Drupal unit testing API being incredibly stupid, the unit tests can only work with PHP >=5.3 while the module will work gracefully with PHP 5.2 (at least using the PhpRedis client). I did not find any hint about making tests being configurable, so per default the tested Redis server must always be on localhost with default configuration.