RedisDB - Perl extension to access redis database


    use RedisDB;

    my $redis = RedisDB->new(host => 'localhost', port => 6379);
    $redis->set($key, $value);
    my $value = $redis->get($key);


This module provides interface to access redis key-value store, it transparently handles disconnects and forks, supports transactions, pipelining, and subscription mode.



Creates a new RedisDB object. The following options are accepted:


domain name of the host running redis server. Default: "localhost"


port to connect. Default: 6379


you can connect to redis using UNIX socket. In this case instead of "host" and "port" you should specify path.


Password, if redis server requires authentication. Alternatively you can use auth method after creating the object.


DB number to use. Specified database will be selected immediately after connecting to the server. Database changes when you sending select command to the server. You can get current database using selected_database method. Default value is 0.


A Redis URL as described in URI::redis.

You cannot use url together with any of host, port, path, password, database.


By default if redis-server returned error reply, or there was a connection error get_reply method throws an exception of RedisDB::Error type, if you set this parameter to false it will return an error object instead. Note, that if you set this to false you should always check if the result you've got from RedisDB method is a RedisDB::Error object.


IO timeout. With this option set, if I/O operation has taken more than specified number of seconds, module will croak or return RedisDB::Error::EAGAIN error object depending on "raise_error" setting. Note, that some OSes do not support SO_RCVTIMEO, and SO_SNDTIMEO socket options, in this case timeout will not work.


Assume that all data on the server encoded in UTF-8. As result all strings will be converted to UTF-8 before sending to server, and all results will be decoded from UTF-8. See "UTF-8 SUPPORT".


After establishing a connection set its name to the specified using "CLIENT SETNAME" command.


by default new establishes a connection to the server. If this parameter is set, then connection will be established only when you will send first command to the server.


this parameter allows you to specify how many attempts to (re)connect to the server should be made before returning an error. Default value is 1, set to -1 if module should try to reconnect indefinitely.


module waits some time before every new attempt to connect. Delay increases each time. This parameter allows you to specify maximum delay between attempts to reconnect. Default value is 10.


if module failed to establish connection with the server it will invoke this callback. First argument to the callback is a reference to the RedisDB object, and second is the error description. You must not invoke any methods on the object inside the callback, but you can change port and host, or path attributes of the RedisDB object to point to another server. After callback returned, module tries to establish connection again using new parameters. To prevent further connection attempts callback should throw an exception, which is done by default callback. This may be useful to switch to backup server if primary went down. RedisDB distribution includes an example of using this callback in eg/

$self->execute($command, @arguments)

send a command to the server, wait for the result and return it. It will throw an exception if the server returns an error or return RedisDB::Error object depending on "raise_error" parameter. It may be more convenient to use instead of this method wrapper named after the corresponding redis command. E.g.:

    $redis->execute('set', key => 'value');
    # is the same as
    $redis->set(key => 'value');

See "WRAPPER METHODS" section for the full list of defined aliases.

Note, that you can not use execute if you have sent some commands using send_command method without the callback argument and have not yet got all replies.

$self->send_command($command[, @arguments][, \&callback])

send a command to the server. If send has failed command will die or return RedisDB::Error object depending on raise_error parameter. Note, that it does not return reply from the server, if callback was not specified, you should retrieve result using get_reply method, otherwise callback will be invoked upon receiving the result with two arguments: the RedisDB object, and the reply from the server. If the server returned an error, the second argument to the callback will be a RedisDB::Error object, you can get description of the error using this object in string context. If you are not interested in reply, you can use RedisDB::IGNORE_REPLY constant as the last argument.

Note, that RedisDB does not run any background threads, so it will not receive the reply and invoke the callback unless you call some of its methods which check if there are replies from the server, like send_command, reply_ready, get_reply, or get_all_replies.


this method may be used in the pipelining mode to check if there are some replies already received from the server. Returns true if there are replies ready to be fetched with get_reply method.


this method blocks till all replies from the server will be received. Note, that callbacks for some replies may send new requests to the server and so this method may block for indefinite time.


receive and return reply from the server. If the server returned an error, method throws RedisDB::Error exception or returns RedisDB::Error object, depending on the "raise_error" parameter.


wait till replies to all the commands without callback set will be received. Returns a list of replies to these commands. For commands with callback set replies are processed as usual. Unlike mainloop this method blocks only till replies to all commands for which callback was NOT set will be received.


return the number of commands sent to the server replies to which were not yet retrieved with get_reply or get_all_replies. This number only includes commands for which callback was not set.


get currently selected database.


reset connection. This method closes existing connection and drops all previously sent requests. After invoking this method the object returns to the same state as it was returned by the constructor.


return the version of the server the client is connected to. The version is returned as a floating point number represented the same way as the perl versions. E.g. for redis 2.1.12 it will return 2.001012.


Instead of using execute and send_command methods directly, it may be more convenient to use wrapper methods with names matching names of the redis commands. These methods call execute or send_command depending on the presence of the callback argument. If callback is specified, the method invokes send_command and returns as soon as the command has been sent to the server; when the reply is received, it will be passed to the callback (see "PIPELINING SUPPORT"). If there is no callback, the method invokes execute, waits for the reply from the server, and returns it. E.g.:

    $val = $redis->get($key);
    # equivalent to
    $val = $redis->execute("get", $key);

    $redis->get($key, sub { $val = $_[1] });
    # equivalent to
    $redis->send_command("get", $key, sub { $val = $_[1] });

The following wrapper methods are defined: append, asking, auth, bgrewriteaof, bgsave, bitcount, bitop, bitpos, blpop, brpop, brpoplpush, client, client_kill, client_getname, client_setname, cluster, command, config, config_get, config_set, config_resetstat, config_rewrite, dbsize, debug_error, debug_object, debug_segfault, decr, decrby, del, dump, echo, eval, evalsha, exists, expire, expireat, flushall, flushdb, geoadd, geodist, geohash, geopos, georadius, georadiusbymember, get, getbit, getrange, getset, hdel, hexists, hget, hgetall, hincrby, hincrbyfloat, hkeys, hlen, hmget, hscan, hmset, hset, hsetnx, hvals, incr, incrby, incrbyfloat, keys, lastsave, lindex, linsert, llen, lpop, lpush, lpushx, lrange, lrem, lset, ltrim, mget, migrate, move, mset, msetnx, object, object_refcount, object_encoding, object_idletime, persist, pexpire, pexpireat, pfadd, pfcount, pfmerge, ping, psetex, pttl, publish, pubsub, pubsub_channels, pubsub_numsub, pubsub_numpat, quit, randomkey, rename, renamenx, restore, rpop, rpoplpush, rpush, rpushx, sadd, save, scan, scard, script, script_exists, script_flush, script_kill, script_load, sdiff, sdiffstore, select, set, setbit, setex, setnx, setrange, sinter, sinterstore, sismember, slaveof, slowlog, smembers, smove, sort, spop, srandmember, srem, sscan strlen, sunion, sunionstore, time, ttl, type, unwatch, watch, zadd, zcard, zcount, zincrby, zinterstore, zlexcount, zrange, zrangebylex, zrangebyscore, zrank, zrem, zremrangebylex, zremrangebyrank, zremrangebyscore, zrevrange, zrevrangebyscore, zrevrank, zscan, zscore, zunionstore.

See description of all commands in redis documentation at

The following commands implement some additional postprocessing of the results:


return information and statistics about the server. Redis-server returns information in form of field:value, the info method parses result and returns it as a hash reference.


return list of clients connected to the server. This method parses server output and returns result as reference to array of hashes.


return information and statistics about the cluster. Redis-server returns information in form of field:value, the cluster_info method parses result and returns it as a hash reference.


return list of cluster nodes. Each node represented as a hash with the following keys: node_id, address, host, port, flags, master_id, last_ping_sent, last_pong_received, link_state, slots.


return reference to a hash describing the role of the server. Hash contains "role" element that can be either "master", "slave", or "sentinel". For master hash will also contain "replication_offset" and "slaves" elements, for slave it will contain "master", "status", and "replication_offset" elements, and for sentinel it will contain "services".


Shuts the redis server down. Returns undef, as the server doesn't send the answer. Croaks in case of the error.

$self->scan_all([MATCH => $pattern,][COUNT => $count,])

this method starts a new SCAN iteration and executes SCAN commands till cursor returned by server is 0. It then returns all the keys returned by server during the iteration. MATCH and COUNT are passed to SCAN command. In case of success returns reference to array with matching keys, in case of error dies or returns RedisDB::Error object depending on raise_error option.

$self->hscan_all($key, [MATCH => $pattern,][COUNT => $count,])

$self->sscan_all($key, [MATCH => $pattern,][COUNT => $count,])

$self->zscan_all($key, [MATCH => $pattern,][COUNT => $count,])

these three methods are doing the same thing as scan_all except that they require a key as the first parameter, and they iterate using HSCAN, SSCAN and ZSCAN commands.


The redis protocol is designed to work with the binary data, both keys and values are encoded in the same way as sequences of octets. By default this module expects all data to be just strings of bytes. There is an option to treat all data as UTF-8 strings. If you pass utf8 parameter to the constructor, module will encode all strings to UTF-8 before sending them to server, and will decode all strings received from server from UTF-8. This has following repercussions you should be aware off: first, you can't store binary data on server with this option on, it would be treated as a sequence of latin1 characters, and would be converted into a corresponding sequence of UTF-8 encoded characters; second, if data returned by the server is not a valid UTF-8 encoded string, the module will croak, and you will have to reinitialize the connection. The parser only checks for invalid UTF-8 byte sequences, it doesn't check if input contains invalid code points. Generally, using this option is not recommended.


If "raise_error" parameter was set to true in the constructor (which is default setting), then module will throw an exception in case network IO function returned an error, or if redis-server returned an error reply. Network exceptions belong to RedisDB::Error::EAGAIN or RedisDB::Error::DISCONNECTED class, if redis-server returned an error exception will be of RedisDB::Error class. If the object was in subscription mode, you will have to restore all the subscriptions. If the object was in the middle of transaction, when after network error you will have to start the transaction again.

If "raise_error" parameter was disabled, then instead of throwing an exception, module will return exception object and also pass this exception object to every callback waiting for the reply from the server. If the object is in subscription mode, then module will automatically restore all subscriptions after reconnect. Note, that during transaction "raise_error" is always enabled, so any error will throw an exception.


Redis server may close a connection if it was idle for some time, also the connection may be closed in case when redis-server was restarted, or just because of the network problem. RedisDB always tries to restore connection to the server if no data has been lost as a result of disconnect, and if "raise_error" parameter disabled it will try to reconnect even if disconnect happened during data transmission. E.g. if the client was idle for some time and the redis server closed the connection, it will be transparently restored when you send a command next time no matter if "raise_error" enabled or not. If you sent a command and the server has closed the connection without sending a complete reply, then module will act differently depending on "raise_error" value. If "raise_error" enabled, the module will cancel all current callbacks, reset the object to the initial state, and throw an exception of RedisDB::Error::DISCONNECTED class, next time you use the object it will establish a new connection. If "raise_error" disabled, the module will pass RedisDB::Error::DISCONNECTED object to all outstanding callbacks and will try to reconnect to the server; it will also automatically restore subscriptions if object was in subscription mode. Module never tries to reconnect after MULTI or WATCH command was sent to server and before corresponding UNWATCH, EXEC or DISCARD was sent as this may cause data corruption, so during transaction module behaves like if "raise_error" is set.

Module makes several attempts to reconnect each time increasing interval before the next attempt, depending on the values of "reconnect_attempts" and "reconnect_delay_max". After each failed attempt to connect module will invoke "on_connect_error" callback which for example may change redis-server hostname, so on next attempt module will try to connect to different server.


You can send commands in the pipelining mode. It means you are sending multiple commands to the server without waiting for the replies. This is implemented by the send_command method. Recommended way of using it is to pass a reference to the callback function as the last argument. When module receives reply from the server, it will call this function with two arguments: reference to the RedisDB object, and reply from the server. It is important to understand though, that RedisDB does not run any background threads, neither it checks for the replies by setting some timer, so e.g. in the following example callback will never be invoked:

    my $pong;
    $redis->send_command( "ping", sub { $pong = $_[1] } );
    sleep 1 while not $pong;    # this will never return

Therefore you need periodically trigger check for the replies. The check is triggered when you call the following methods: send_command, reply_ready, get_reply, get_all_replies. Calling wrapper method, like $redis->get('key'), will also trigger check as internally wrapper methods use methods listed above.

Also you can omit callback argument when invoke send_command. In this case you have to fetch reply later explicitly using get_reply method. This is how synchronous execute is implemented, basically it is:

    sub execute {
        my $self = shift;
        return $self->get_reply;

That is why it is not allowed to call execute unless you have got replies to all commands sent previously with send_command without callback. Using send_command without callback is not recommended.

Sometimes you are not interested in replies sent by the server, e.g. SET command usually just return 'OK', in this case you can pass to send_command callback which ignores its arguments, or use RedisDB::IGNORE_REPLY constant, it is a no-op function:

    for (@keys) {
        # execute will not just send 'GET' command to the server,
        # but it will also receive response to the 'SET' command sent on
        # the previous loop iteration
        my $val = $redis->execute( "get", $_ );
        $redis->send_command( "set", $_, fun($val), RedisDB::IGNORE_REPLY );
    # and this will wait for the last reply

or using "WRAPPER METHODS" you can rewrite it as:

    for (@keys) {
        my $val = $redis->get($_);
        $redis->set( $_, fun($val), RedisDB::IGNORE_REPLY );


RedisDB supports subscriptions to redis channels. In the subscription mode you can subscribe to some channels and receive all the messages sent to these channels. You can subscribe to channels and then manually check messages using get_reply method, or you can invoke subscription_loop method, which will block in loop waiting for messages and invoking callback for each received message. In the first case you can use subscribe and psubscribe methods to subscribe to channels and then you can use get_reply method to get messages from the channel:

        foo => sub {
            my ( $redis, $channel, $patern, $message ) = @_;
            print "Foo: $message\n";
    # Wait for messages
    $res = $redis->get_reply;

get_reply method for messages from the channel will invoke callback specified as the second optional argument of the subscribe method and will also return raw replies from the server, both for messages from the channels and for informational messages from the redis server. If you do not want to block in get_reply method, you can check if there are any messages using reply_ready method.

In the second case you invoke subscription_loop method, it subscribes to specified channels and waits for messages, when a message arrived it invokes callback defined for the channel from which the message came. Here is an example:

    my $message_cb = sub {
        my ( $redis, $channel, $pattern, $message ) = @_;
        print "$channel: $message\n";

    my $control_cb = sub {
        my ( $redis, $channel, $pattern, $message ) = @_;
        if ( $channel eq 'control.quit' ) {
        elsif ( $channel eq 'control.subscribe' ) {

        subscribe        => [ 'news', ],
        psubscribe       => [ 'control.*' => $control_cb ],
        default_callback => $message_cb,

subscription_loop will subscribe you to the "news" channel and "control.*" channels. It will call specified callbacks every time a new message received. When message came from "control.subscribe" channel, callback subscribes to an additional channel. When message came from "control.quit" channel, callback unsubscribes from all channels.

Callbacks used in subscription mode receive four arguments: the RedisDB object, the channel from which the message came, the pattern if you subscribed to this channel using psubscribe method, and the message itself.

Once you switched into subscription mode using either subscribe or psubscribe command, or by entering subscription_loop, you only can send subscribe, psubscribe, unsubscribe, and punsubscribe commands to the server, other commands will throw an exception.

You can publish messages into the channels using the publish method. This method should be called when you in the normal mode, and can't be used while you're in the subscription mode.

Following methods can be used in subscription mode:


Enter into the subscription mode. The method subscribes you to the specified channels, waits for the messages, and invokes the appropriate callback for every received message. The method returns after you unsubscribed from all the channels. It accepts the following parameters:


reference to the default callback. This callback is invoked for a message if you didn't specify other callback for the channel this message comes from.


an array reference. Contains the list of channels you want to subscribe. A channel name may be optionally followed by the reference to a callback function for this channel. E.g.:

    [ 'news', 'messages', 'errors' => \&error_cb, 'other' ]

channels "news", "messages", and "other" will use default callback, but for the "errors" channel error_cb function will be used.


same as subscribe, but you specify patterns for channels' names.

All parameters are optional, but you must subscribe at least to one channel. Also if default_callback is not specified, you have to explicitly specify a callback for every channel you are going to subscribe.

$self->subscribe($channel[, \&callback])

Subscribe to the $channel. If $callback is not specified, default callback will be used in subscription loop, or messages will be returned by get_reply if you are not using subscription loop.

$self->psubscribe($pattern[, \&callback])

Subscribe to channels matching $pattern. If $callback is not specified, default callback will be used in subscription loop, or messages will be returned by get_reply if you are not using subscription loop.


Unsubscribe from the listed @channels. If no channels was specified, unsubscribe from all the channels to which you have subscribed using subscribe.


Unsubscribe from the listed @patterns. If no patterns was specified, unsubscribe from all the channels to which you have subscribed using psubscribe.


Return list of channels to which you have subscribed using subscribe


Return list of channels to which you have subscribed using psubscribe


Transactions allow you to execute a sequence of commands in a single step. In order to start a transaction you should use the multi method. After you have entered a transaction all the commands you issue are queued, but not executed till you call the exec method. Typically these commands return string "QUEUED" as a result, but if there is an error in e.g. number of arguments, they may return an error. When you call exec, all the queued commands will be executed and exec will return a list of results for every command in the transaction. If instead of exec you call discard, all scheduled commands will be canceled.

You can set some keys as watched. If any watched key has been changed by another client before you called exec, the transaction will be discarded and exec will return false value.

$self->watch(@keys[, \&callback])

mark given keys to be watched


unwatch all keys


Enter the transaction. After this and till exec or discard will be called, all the commands will be queued but not executed.


Execute all queued commands and finish the transaction. Returns a list of results for every command. Will croak if some command has failed. Also unwatches all the keys. If some of the watched keys has been changed by other client, the transaction will be canceled and exec will return false.


Discard all queued commands without executing them and unwatch all keys.


For accessing redis cluster use RedisDB::Cluster package


For accessing redis servers managed by sentinel use RedisDB::Sentinel package


Redis, Redis::hiredis, Redis::Client, AnyEvent::Redis, AnyEvent::Redis::RipeRedis


I was in need of the client for redis database. AnyEvent::Redis didn't suite me as it requires an event loop, and it didn't fit into the existing code. The problem with Redis is that it didn't (at the time I started this) reconnect to the server if connection was closed after timeout or as result of the server restart, and it does not support pipelining. After analyzing what I need to change in Redis in order to get all I want, I decided that it will be simpler to write the new module from scratch. This also solves the problem of backward compatibility.


Please report any bugs or feature requests via GitHub bug tracker at

Known bugs are:

Timeout support is OS dependent. If OS doesn't support SO_SNDTIMEO and SO_RCVTIMEO options timeouts will not work.

QUIT command doesn't work with redis-server before version 2.0


Sanko Robinson and FunkyMonk helped me with porting this module to Windows.

HIROSE Masaake fixed handling of commands containing space (like "CONFIG GET")


Pavel Shaydo, <zwon at>


Copyright 2011-2021 Pavel Shaydo.

This program is free software; you can redistribute it and/or modify it under the terms of either: the GNU General Public License as published by the Free Software Foundation; or the Artistic License.

See for more information.