jackrobison/txredisapi

non-blocking redis client for python twisted

txredisapi

For the latest source code, see http://github.com/fiorix/txredisapi

txredisapi is a non-blocking client driver for the redis database, written in Python. It uses Twisted for the asynchronous communication with redis.

It started as a fork of the original redis protocol for twisted, and evolved into a more robust, reliable, and complete solution for applications like web servers. These types of applications often need a fault-tolerant pool of connections with multiple redis servers, making it possible to easily develop and maintain distributed systems.

Most of the redis commands are supported, as well as other features such as silent reconnection, connection pools, and automatic sharding.

This driver is distributed as part of the cyclone web framework.

Changelog

See CHANGELOG.md

Features

• Connection Pools
• Lazy Connections
• Automatic Sharding
• Automatic Reconnection
• Connection using Redis Sentinel
• Publish/Subscribe (PubSub)
• Transactions
• Unix Socket Connections

Install

Bear in mind that txredisapi.py is pure-python, in a single file. Thus, there's absolutely no need to install it. Instead, just copy it to your project directory and start using.

Latest source code is at https://github.com/fiorix/txredisapi.

If you have cyclone, you probably already have it too. Try the following:

$ python
>>> import cyclone.redis
>>> cyclone.redis.version
'1.0'

However, if you really really insist in installing, get it from pypi:

pip install txredisapi

Unit Tests

Twisted Trial unit tests are available. Just start redis, and run trial ./tests. If unix sockets are disabled in redis, it will silently skip those tests.

Make sure you run redis-cli flushall to clean up redis after the tests.

Usage

First thing to do is choose what type of connection you want. The driver supports single connection, connection pools, sharded connections (with automatic distribution based on a built-in consistent hashing algorithm), sharded connection pools, and all of these different types can be lazy, which is explained later (because I'm lazy now).

Basically, you want normal connections for simple batch clients that connect to redis, execute a couple of commands and disconnect - like crawlers, etc.

Example:

#!/usr/bin/env python
# coding: utf-8

import txredisapi as redis

from twisted.internet import defer
from twisted.internet import reactor


@defer.inlineCallbacks
def main():
    rc = yield redis.Connection()
    print rc

    yield rc.set("foo", "bar")
    v = yield rc.get("foo")
    print "foo:", repr(v)

    yield rc.disconnect()


if __name__ == "__main__":
    main().addCallback(lambda ign: reactor.stop())
    reactor.run()

Easily switch between redis.Connection() and redis.ConnectionPool() with absolutely no changes to the logic of your program.

These are all the supported methods for connecting to Redis::

Connection(host, port, dbid, reconnect, charset)
lazyConnection(host, port, dbid, reconnect, charset)

ConnectionPool(host, port, dbid, poolsize, reconnect, charset)
lazyConnectionPool(host, port, dbid, poolsize, reconnect, charset)

ShardedConnection(hosts, dbid, reconnect, charset)
lazyShardedConnection(hosts, dbid, reconnect, charset)

ShardedConnectionPool(hosts, dbid, poolsize, reconnect, charset)
lazyShardedConnectionPool(hosts, dbid, poolsize, reconnect, charset)

UnixConnection(path, dbid, reconnect, charset)
lazyUnixConnection(path, dbid, reconnect, charset)

UnixConnectionPool(unix, dbid, poolsize, reconnect, charset)
lazyUnixConnectionPool(unix, dbid, poolsize, reconnect, charset)

ShardedUnixConnection(paths, dbid, reconnect, charset)
lazyShardedUnixConnection(paths, dbid, reconnect, charset)

ShardedUnixConnectionPool(paths, dbid, poolsize, reconnect, charset)
lazyShardedUnixConnectionPool(paths, dbid, poolsize, reconnect, charset)

The arguments are:

• host: the IP address or hostname of the redis server. [default: localhost]
• port: port number of the redis server. [default: 6379]
• path: path of redis server's socket [default: /tmp/redis.sock]
• dbid: database id of redis server. [default: 0]
• poolsize: how many connections to make. [default: 10]
• reconnect: auto-reconnect if connection is lost. [default: True]
• charset: string encoding. Do not decode/encode strings if None. [default: utf-8]
• hosts (for sharded): list of host:port pairs. [default: None]
• paths (for sharded): list of pathnames. [default: None]
• password: password for the redis server. [default: None]

Connection Handlers

All connection methods return a connection handler object at some point.

Normal connections (not lazy) return a deferred, which is fired with the connection handler after the connection is established.

In case of connection pools, it will only fire the callback after all connections are set up, and ready.

Connection handler is the client interface with redis. It accepts all the commands supported by redis, such as get, set, etc. It is the rc object in the example below.

Connection handlers will automatically select one of the available connections in connection pools, and automatically reconnect to redis when necessary.

If the connection with redis is lost, all commands will raise the ConnectionError exception, to indicate that there's no active connection. However, if the reconnect argument was set to True during the initialization, it will continuosly try to reconnect, in background.

Example:

#!/usr/bin/env python
# coding: utf-8

import txredisapi as redis

from twisted.internet import defer
from twisted.internet import reactor


def sleep(n):
    d = defer.Deferred()
    reactor.callLater(5, lambda *ign: d.callback(None))
    return d


@defer.inlineCallbacks
def main():
    rc = yield redis.ConnectionPool()
    print rc

    # set
    yield rc.set("foo", "bar")

    # sleep, so you can kill redis
    print "sleeping for 5s, kill redis now..."
    yield sleep(5)

    try:
      v = yield rc.get("foo")
      print "foo:", v

      yield rc.disconnect()
    except redis.ConnectionError, e:
      print str(e)


if __name__ == "__main__":
    main().addCallback(lambda ign: reactor.stop())
    reactor.run()

Lazy Connections

This type of connection will immediately return the connection handler object, even before the connection is made.

It will start the connection, (or connections, in case of connection pools) in background, and automatically reconnect if necessary.

You want lazy connections when you're writing servers, like web servers, or any other type of server that should not wait for the redis connection during the initialization of the program.

The example below is a web application, which will expose redis set, get and delete commands over HTTP.

If the database connection is down (either because redis is not running, or whatever reason), the web application will start normally. If connection is lost during the operation, nothing will change.

When there's no connection, all commands will fail, therefore the web application will respond with HTTP 503 (Service Unavailable). It will resume to normal once the connection with redis is re-established.

Try killing redis server after the application is running, and make a couple of requests. Then, start redis again and give it another try.

Example:

#!/usr/bin/env python
# coding: utf-8

import sys

import cyclone.web
import cyclone.redis
from twisted.internet import defer
from twisted.internet import reactor
from twisted.python import log


class Application(cyclone.web.Application):
    def __init__(self):
      handlers = [ (r"/text/(.+)", TextHandler) ]

      RedisMixin.setup()
      cyclone.web.Application.__init__(self, handlers, debug=True)


class RedisMixin(object):
    redis_conn = None

    @classmethod
    def setup(self):
        RedisMixin.redis_conn = cyclone.redis.lazyConnectionPool()


# Provide GET, SET and DELETE redis operations via HTTP
class TextHandler(cyclone.web.RequestHandler, RedisMixin):
    @defer.inlineCallbacks
    def get(self, key):
      try:
          value = yield self.redis_conn.get(key)
      except Exception, e:
          log.msg("Redis failed to get('%s'): %s" % (key, str(e)))
          raise cyclone.web.HTTPError(503)

      self.set_header("Content-Type", "text/plain")
      self.write("%s=%s\r\n" % (key, value))

    @defer.inlineCallbacks
    def post(self, key):
        value = self.get_argument("value")
        try:
            yield self.redis_conn.set(key, value)
        except Exception, e:
            log.msg("Redis failed to set('%s', '%s'): %s" % (key, value, str(e)))
            raise cyclone.web.HTTPError(503)

        self.set_header("Content-Type", "text/plain")
        self.write("%s=%s\r\n" % (key, value))

    @defer.inlineCallbacks
    def delete(self, key):
        try:
            n = yield self.redis_conn.delete(key)
        except Exception, e:
            log.msg("Redis failed to del('%s'): %s" % (key, str(e)))
            raise cyclone.web.HTTPError(503)

        self.set_header("Content-Type", "text/plain")
        self.write("DEL %s=%d\r\n" % (key, n))


def main():
    log.startLogging(sys.stdout)
    reactor.listenTCP(8888, Application(), interface="127.0.0.1")
    reactor.run()


if __name__ == "__main__":
    main()

This is the server running in one terminal::

$ ./helloworld.py
2012-02-17 15:40:25-0500 [-] Log opened.
2012-02-17 15:40:25-0500 [-] Starting factory <redis.Factory instance at 0x1012f0560>
2012-02-17 15:40:25-0500 [-] __main__.Application starting on 8888
2012-02-17 15:40:25-0500 [-] Starting factory <__main__.Application instance at 0x100f42290>
2012-02-17 15:40:53-0500 [RedisProtocol,client] 200 POST /text/foo (127.0.0.1) 1.20ms
2012-02-17 15:41:01-0500 [RedisProtocol,client] 200 GET /text/foo (127.0.0.1) 0.97ms
2012-02-17 15:41:09-0500 [RedisProtocol,client] 200 DELETE /text/foo (127.0.0.1) 0.65ms
(killed redis-server)
2012-02-17 15:48:48-0500 [HTTPConnection,0,127.0.0.1] Redis failed to get('foo'): Not connected
2012-02-17 15:48:48-0500 [HTTPConnection,0,127.0.0.1] 503 GET /text/foo (127.0.0.1) 2.99ms

And these are the requests, from curl in another terminal.

Set:

$ curl -D - -d "value=bar" http://localhost:8888/text/foo
HTTP/1.1 200 OK
Content-Length: 9
Content-Type: text/plain

foo=bar

Get:

$ curl -D - http://localhost:8888/text/foo
HTTP/1.1 200 OK
Content-Length: 9
Etag: "b63729aa7fa0e438eed735880951dcc21d733676"
Content-Type: text/plain

foo=bar

Delete:

$ curl -D - -X DELETE http://localhost:8888/text/foo
HTTP/1.1 200 OK
Content-Length: 11
Content-Type: text/plain

DEL foo=1

When redis is not running:

$ curl -D - http://localhost:8888/text/foo
HTTP/1.1 503 Service Unavailable
Content-Length: 89
Content-Type: text/html; charset=UTF-8

<html><title>503: Service Unavailable</title>
<body>503: Service Unavailable</body></html>

Sharded Connections

They can be normal, or lazy connections. They can be sharded connection pools. Not all commands are supported on sharded connections.

If the command you're trying to run is not supported on sharded connections, the connection handler will raise the NotImplementedError exception.

Simple example with automatic sharding of keys between two redis servers:

#!/usr/bin/env python
# coding: utf-8

import txredisapi as redis

from twisted.internet import defer
from twisted.internet import reactor


@defer.inlineCallbacks
def main():
    rc = yield redis.ShardedConnection(["localhost:6379", "localhost:6380"])
    print rc
    print "Supported methods on sharded connections:", rc.ShardedMethods

    keys = []
    for x in xrange(100):
        key = "foo%02d" % x
        yield rc.set(key, "bar%02d" % x)
        keys.append(key)

    # yey! mget is supported!
    response = yield rc.mget(keys)
    for val in response:
        print val

    yield rc.disconnect()


if __name__ == "__main__":
    main().addCallback(lambda ign: reactor.stop())
    reactor.run()

Transactions

For obvious reasons, transactions are NOT supported on sharded connections. But they work pretty good on normal or lazy connections, and connection pools.

NOTE: redis uses the following methods for transactions:

• WATCH: synchronization
• MULTI: start the transaction
• EXEC: commit the transaction
• DISCARD: you got it.

Because exec is a reserved word in Python, the command to commit is commit.

Example:

#!/usr/bin/env python
# coding: utf-8

import txredisapi as redis

from twisted.internet import defer
from twisted.internet import reactor


@defer.inlineCallbacks
def main():
    rc = yield redis.ConnectionPool()

    # Remove the keys
    yield rc.delete(["a1", "a2", "a3"])

    # Start transaction
    t = yield rc.multi()

    # These will return "QUEUED" - even t.get(key)
    yield t.set("a1", "1")
    yield t.set("a2", "2")
    yield t.set("a3", "3")
    yield t.get("a1")

    # Try to call get() while in a transaction.
    # It will fail if it's not a connection pool, or if all connections
    # in the pool are in a transaction.
    # Note that it's rc.get(), not the transaction object t.get().
    try:
        v = yield rc.get("foo")
    print "foo=", v
        except Exception, e:
        print "can't get foo:", e

    # Commit, and get all responses from transaction.
    r = yield t.commit()
    print "commit=", repr(r)

    yield rc.disconnect()


if __name__ == "__main__":
    main().addCallback(lambda ign: reactor.stop())
    reactor.run()

A "COUNTER" example, using WATCH/MULTI:

 #!/usr/bin/env python
 # coding: utf-8

 import txredisapi as redis

 from twisted.internet import defer
 from twisted.internet import reactor


 @defer.inlineCallbacks
 def main():
     rc = yield redis.ConnectionPool()

     # Reset keys
     yield rc.set("a1", 0)

     # Synchronize and start transaction
     t = yield rc.watch("a1")

     # Load previous value
     a1 = yield t.get("a1")

     # start the transactional pipeline
     yield t.multi()

     # modify and retrieve the new a1 value
     yield t.set("a1", a1 + 1)
     yield t.get("a1")

     print "simulating concurrency, this will abort the transaction"
     yield rc.set("a1", 2)

     try:
         r = yield t.commit()
         print "commit=", repr(r)
     except redis.WatchError, e:
         a1 = yield rc.get("a1")
         print "transaction has failed."
         print "current a1 value: ", a1

     yield rc.disconnect()


 if __name__ == "__main__":
     main().addCallback(lambda ign: reactor.stop())
     reactor.run()

Calling commit will cause it to return a list with the return of all commands executed in the transaction. discard, on the other hand, will normally return just an OK.

Pipelining

txredisapi automatically pipelines all commands by sending next commands without waiting for the previous one to receive reply from server. This works even on single connections and increases performance by reducing number of round-trip delays and. There are two exceptions, though:

• no commands will be sent after blocking blpop, brpop or brpoplpush until response is received;
• transaction by multi/commit are also blocking connection making all other commands to wait until transaction is executed.

When you need to load tons of data to Redis it might be more effective to sent commands in batches grouping them together offline to save on TCP packets and network stack overhead. You can do this using pipeline method to explicitly accumulate commands and send them to server in a single batch. Be careful to not accumulate too many commands: unreasonable batch size may eat up unexpected amount of memory on both client and server side. Group commands in batches of, for example, 10k commands instead of sending all your data at once. The speed will be nearly the same, but the additional memory used will be at max the amount needed to queue this 10k commands

To send commands in a batch:

#!/usr/bin/env python
# coding: utf-8

import txredisapi as redis

from twisted.internet import defer
from twisted.internet import reactor

@defer.inlineCallbacks
def main():
    rc = yield redis.ConnectionPool()

    # Start grouping commands
    pipeline = yield rc.pipeline()

    pipeline.set("foo", 123)
    pipeline.set("bar", 987)
    pipeline.get("foo")
    pipeline.get("bar")

    # Write those 2 sets and 2 gets to redis all at once, and wait
    # for all replies before continuing.
    results = yield pipeline.execute_pipeline()

    print "foo:", results[2] # should be 123
    print "bar:", results[3] # should be 987

    yield rc.disconnect()

if __name__ == "__main__":
    main().addCallback(lambda ign: reactor.stop())
    reactor.run()

Authentication

This is how to authenticate::

#!/usr/bin/env python

import txredisapi
from twisted.internet import defer
from twisted.internet import reactor


@defer.inlineCallbacks
def main():
    redis = yield txredisapi.Connection(password="foobared")
    yield redis.set("foo", "bar")
    print (yield redis.get("foo"))
    reactor.stop()


if __name__ == "__main__":
    main()
    reactor.run()

Connection using Redis Sentinel

txredisapi can discover Redis master and slaves addresses using Redis Sentinel and automatically failover in case of server failure.

#!/usr/bin/env python

from twisted.internet.task import react
import txredisapi

@defer.inlineCallbacks
def main(reactor):
    sentinel = txredisapi.Sentinel([("sentinel-a", 26379), ("sentinel-b", 26379), ("sentinel-c", 26379)])
    redis = sentinel.master_for("service_name")
    yield redis.set("foo", "bar")
    print (yield redis.get("foo"))
    yield redis.disconnect()
    yield sentinel.disconnect()
    
react(main)

Usual connection arguments like dbid=N or poolsize=N can be specified in master_for() call. Use sentinel.slave_for() to connect to one of the slaves instead of master.

Add min_other_sentinels=N to Sentinel constructor call to make it obey information only from sentinels that currently connected to specified number of other sentinels to minimize a risk of split-brain in case of network partitioning.

Credits

Thanks to (in no particular order):

• Gleicon Moraes

  • Bug fixes, testing, and RestMQ.
  • For writing the Consistent Hashing algorithm used for sharding.

• Dorian Raymer and Ludovico Magnocavallo

  • Authors of the original redis protocol for twisted.

• Vanderson Mota

  • Initial pypi setup, and patches.

• Jeethu Rao

  • Contributed with test cases, and other ideas like support for travis-ci

• Jeremy Archer

  • Minor bugfixes.

• Christoph Tavan (@ctavan)

  • Idea and test case for nested multi bulk replies, minor command enhancements.

• dgvncsz0f

  • WATCH/UNWATCH commands

• Ilia Glazkov

  • Free connection selection algorithm for pools.
  • Non-unicode charset fixes.
  • SCAN commands

• Matt Pizzimenti (mjpizz)

  • pipelining support

• Nickolai Novik (jettify)

  • update of SET command

• Evgeny Tataurov (etataurov)

  • Ability to use hiredis protocol parser

• Ilya Skriblovsky (IlyaSkriblovsky)

  • Sentinel support