Skip to content
forked from jamesls/fakeredis

Fake implementation of redis API (redis-py) for testing purposes

License

Notifications You must be signed in to change notification settings

jd014/fakeredis

 
 

Repository files navigation

fakeredis: A fake version of a redis-py

https://coveralls.io/repos/jamesls/fakeredis/badge.svg?branch=master

fakeredis is a pure-Python implementation of the redis-py python client that simulates talking to a redis server. This was created for a single purpose: to write unittests. Setting up redis is not hard, but many times you want to write unittests that do not talk to an external server (such as redis). This module now allows tests to simply use this module as a reasonable substitute for redis.

Although fakeredis is pure Python, you will need lupa if you want to run Lua scripts (this includes features like redis.lock.Lock, which are implemented in Lua). If you install fakeredis with pip install fakeredis[lua] it will be automatically installed.

Alternatives

Consider using redislite instead of fakeredis. It runs a real redis server and connects to it over a UNIX domain socket, so it will behave just like a real server. Another alternative is birdisle, which runs the redis code as a Python extension (no separate process), but which is currently unmaintained.

How to Use

The intent is for fakeredis to act as though you're talking to a real redis server. It does this by storing state internally. For example:

>>> import fakeredis
>>> r = fakeredis.FakeStrictRedis()
>>> r.set('foo', 'bar')
True
>>> r.get('foo')
'bar'
>>> r.lpush('bar', 1)
1
>>> r.lpush('bar', 2)
2
>>> r.lrange('bar', 0, -1)
[2, 1]

The state is stored in an instance of FakeServer. If one is not provided at construction, a new instance is automatically created for you, but you can explicitly create one to share state:

>>> import fakeredis
>>> server = fakeredis.FakeServer()
>>> r1 = fakeredis.FakeStrictRedis(server=server)
>>> r1.set('foo', 'bar')
True
>>> r2 = fakeredis.FakeStrictRedis(server=server)
>>> r2.get('foo')
'bar'
>>> r2.set('bar', 'baz')
True
>>> r1.get('bar')
'baz'
>>> r2.get('bar')
'baz'

It is also possible to mock connection errors so you can effectively test your error handling. Simply set the connected attribute of the server to False after initialization.

>>> import fakeredis
>>> server = fakeredis.FakeServer()
>>> server.connected = False
>>> r = fakeredis.FakeStrictRedis(server=server)
>>> r.set('foo', 'bar')
ConnectionError: FakeRedis is emulating a connection error.
>>> server.connected = True
>>> r.set('foo', 'bar')
True

Fakeredis implements the same interface as redis-py, the popular redis client for python, and models the responses of redis 6.2 (although most new features are not supported).

Support for aioredis

You can also use fakeredis to mock out aioredis. This is a much newer addition to fakeredis (added in 1.4.0) with less testing, so your mileage may vary. Both version 1 and version 2 (which have very different APIs) are supported. The API provided by fakeredis depends on the version of aioredis that is installed.

aioredis 1.x

Example:

>>> import fakeredis.aioredis
>>> r = await fakeredis.aioredis.create_redis_pool()
>>> await r.set('foo', 'bar')
True
>>> await r.get('foo')
b'bar'

You can pass a FakeServer as the first argument to create_redis or create_redis_pool to share state (you can even share state with a fakeredis.FakeRedis). It should even be safe to do this state sharing between threads (as long as each connection/pool is only used in one thread).

It is highly recommended that you only use the aioredis support with Python 3.5.3 or higher. Earlier versions will not work correctly with non-default event loops.

aioredis 2.x

Example:

>>> import fakeredis.aioredis
>>> r = fakeredis.aioredis.FakeRedis()
>>> await r.set('foo', 'bar')
True
>>> await r.get('foo')
b'bar'

The support is essentially the same as for redis-py e.g., you can pass a server keyword argument to the FakeRedis constructor.

Porting to fakeredis 1.0

Version 1.0 is an almost total rewrite, intended to support redis-py 3.x and improve the Lua scripting emulation. It has a few backwards incompatibilities that may require changes to your code:

  1. By default, each FakeRedis or FakeStrictRedis instance contains its own state. This is equivalent to the singleton=False option to previous versions of fakeredis. This change was made to improve isolation between tests. If you need to share state between instances, create a FakeServer, as described above.
  2. FakeRedis is now a subclass of Redis, and similarly FakeStrictRedis is a subclass of StrictRedis. Code that uses isinstance may behave differently.
  3. The connected attribute is now a property of FakeServer, rather than FakeRedis or FakeStrictRedis. You can still pass the property to the constructor of the latter (provided no server is provided).

Unimplemented Commands

All of the redis commands are implemented in fakeredis with these exceptions:

server

  • acl load
  • acl save
  • acl list
  • acl users
  • acl getuser
  • acl setuser
  • acl deluser
  • acl cat
  • acl genpass
  • acl whoami
  • acl log
  • acl help
  • bgrewriteaof
  • command
  • command count
  • command getkeys
  • command info
  • config get
  • config rewrite
  • config set
  • config resetstat
  • debug object
  • debug segfault
  • info
  • lolwut
  • memory doctor
  • memory help
  • memory malloc-stats
  • memory purge
  • memory stats
  • memory usage
  • module list
  • module load
  • module unload
  • monitor
  • role
  • shutdown
  • slaveof
  • replicaof
  • slowlog
  • sync
  • psync
  • latency doctor
  • latency graph
  • latency history
  • latency latest
  • latency reset
  • latency help

connection

  • auth
  • client caching
  • client id
  • client kill
  • client list
  • client getname
  • client getredir
  • client pause
  • client reply
  • client setname
  • client tracking
  • client unblock
  • hello
  • quit

string

  • bitfield
  • bitop
  • bitpos
  • stralgo

sorted_set

  • bzpopmin
  • bzpopmax
  • zpopmax
  • zpopmin

cluster

  • cluster addslots
  • cluster bumpepoch
  • cluster count-failure-reports
  • cluster countkeysinslot
  • cluster delslots
  • cluster failover
  • cluster flushslots
  • cluster forget
  • cluster getkeysinslot
  • cluster info
  • cluster keyslot
  • cluster meet
  • cluster myid
  • cluster nodes
  • cluster replicate
  • cluster reset
  • cluster saveconfig
  • cluster set-config-epoch
  • cluster setslot
  • cluster slaves
  • cluster replicas
  • cluster slots
  • readonly
  • readwrite

generic

  • migrate
  • object
  • touch
  • wait

geo

  • geoadd
  • geohash
  • geopos
  • geodist
  • georadius
  • georadiusbymember

list

  • lpos

pubsub

  • pubsub

scripting

  • script debug
  • script kill

stream

  • xinfo
  • xadd
  • xtrim
  • xdel
  • xrange
  • xrevrange
  • xlen
  • xread
  • xgroup
  • xreadgroup
  • xack
  • xclaim
  • xpending

Other limitations

Apart from unimplemented commands, there are a number of cases where fakeredis won't give identical results to real redis. The following are differences that are unlikely to ever be fixed; there are also differences that are fixable (such as commands that do not support all features) which should be filed as bugs in Github.

  1. Hyperloglogs are implemented using sets underneath. This means that the type command will return the wrong answer, you can't use get to retrieve the encoded value, and counts will be slightly different (they will in fact be exact).
  2. When a command has multiple error conditions, such as operating on a key of the wrong type and an integer argument is not well-formed, the choice of error to return may not match redis.
  3. The incrbyfloat and hincrbyfloat commands in redis use the C long double type, which typically has more precision than Python's float type.
  4. Redis makes guarantees about the order in which clients blocked on blocking commands are woken up. Fakeredis does not honour these guarantees.
  5. Where redis contains bugs, fakeredis generally does not try to provide exact bug-compatibility. It's not practical for fakeredis to try to match the set of bugs in your specific version of redis.
  6. There are a number of cases where the behaviour of redis is undefined, such as the order of elements returned by set and hash commands. Fakeredis will generally not produce the same results, and in Python versions before 3.6 may produce different results each time the process is re-run.
  7. SCAN/ZSCAN/HSCAN/SSCAN will not necessarily iterate all items if items are deleted or renamed during iteration. They also won't necessarily iterate in the same chunk sizes or the same order as redis.
  8. DUMP/RESTORE will not return or expect data in the RDB format. Instead the pickle module is used to mimic an opaque and non-standard format. WARNING: Do not use RESTORE with untrusted data, as a malicious pickle can execute arbitrary code.

Contributing

Contributions are welcome. Please see the contributing guide for more details. The maintainer generally has very little time to work on fakeredis, so the best way to get a bug fixed is to contribute a pull request.

If you'd like to help out, you can start with any of the issues labeled with HelpWanted.

Running the Tests

To ensure parity with the real redis, there are a set of integration tests that mirror the unittests. For every unittest that is written, the same test is run against a real redis instance using a real redis-py client instance. In order to run these tests you must have a redis server running on localhost, port 6379 (the default settings). WARNING: the tests will completely wipe your database!

First install the requirements file:

pip install -r requirements.txt

To run all the tests:

pytest

If you only want to run tests against fake redis, without a real redis:

pytest -m fake

Because this module is attempting to provide the same interface as redis-py, the python bindings to redis, a reasonable way to test this to to take each unittest and run it against a real redis server. fakeredis and the real redis server should give the same result. To run tests against a real redis instance instead:

pytest -m real

If redis is not running and you try to run tests against a real redis server, these tests will have a result of 's' for skipped.

There are some tests that test redis blocking operations that are somewhat slow. If you want to skip these tests during day to day development, they have all been tagged as 'slow' so you can skip them by running:

pytest -m "not slow"

Revision history

1.7.1

  • #324 Support redis-py 4.1.x

1.7.0

  • Change a number of corner-case behaviours to match Redis 6.2.6.
  • #310 Fix DeprecationWarning for sampling from a set
  • #315 Improved support for constructor arguments
  • #316 Support redis-py 4
  • #319 Add support for GET option to SET
  • #323 PERSIST and EXPIRE should invalidate watches

1.6.1

  • #305 Some packaging modernisation
  • #306 Fix FakeRedisMixin.from_url for unix sockets
  • #308 Remove use of async_generator from tests

1.6.0

  • #304 Support aioredis 2
  • #302 Switch CI from Travis CI to Github Actions

1.5.2

  • Depend on aioredis<2 (aioredis 2.x is a backwards-incompatible rewrite).

1.5.1

  • #298 Fix a deadlock caused by garbage collection

1.5.0

  • Fix clearing of watches when a transaction is aborted.
  • Support Python 3.9 and drop support for Python 3.5.
  • Update handling of EXEC failures to match redis 6.0.6+.
  • #293 Align FakeConnection constructor signature to base class
  • Skip hypothesis tests on 32-bit Redis servers.

1.4.5

  • #285 Add support for DUMP and RESTORE commands
  • #286 Add support for TYPE option to SCAN command

1.4.4

  • #281 Add support for SCRIPT EXISTS and SCRIPT FLUSH subcommands
  • #280 Fix documentation about singleton argument

1.4.3

  • #277 Implement SET with KEEPTTL
  • #278 Handle indefinite timeout for PUBSUB commands

1.4.2

  • #269 Prevent passing booleans from Lua to redis
  • #254 Implement TIME command
  • #232 Implement ZADD with INCR
  • Rework of unit tests to use more pytest idioms

1.4.1

  • #268 Support redis-py 3.5 (no code changes, just setup.py)

1.4.0

  • Add support for aioredis.
  • Fix interaction of no-op SREM with WATCH.

1.3.1

  • Make errors from Lua behave more like real redis

1.3.0

  • #266 Implement redis.log in Lua

1.2.1

  • #262 Cannot repr redis object without host attribute
  • Fix a bug in the hypothesis test framework that occasionally caused a failure

1.2.0

  • Drop support for Python 2.7.
  • Test with Python 3.8 and Pypy3.
  • Refactor Hypothesis-based tests to support the latest version of Hypothesis.
  • Fix a number of bugs in the Hypothesis tests that were causing spurious test failures or hangs.
  • Fix some obscure corner cases
    • If a WATCHed key is MOVEd, don't invalidate the transaction.
    • Some cases of passing a key of the wrong type to SINTER/SINTERSTORE were not reporting a WRONGTYPE error.
    • ZUNIONSTORE/ZINTERSTORE could generate different scores from real redis in corner cases (mostly involving infinities).
  • Speed up the implementation of BINCOUNT.

1.1.1

  • Support redis-py 3.4.

1.1.0

  • #257 Add other inputs for redis connection

1.0.5

  • #247 Support NX/XX/CH flags in ZADD command
  • #250 Implement UNLINK command
  • #252 Fix implementation of ZSCAN

1.0.4

1.0.3

  • #235 Support for redis==3.2

1.0.2

  • #235 Depend on redis<3.2

1.0.1

  • Fix crash when a connection closes without unsubscribing and there is a subsequent PUBLISH

1.0

Version 1.0 is a major rewrite. It works at the redis protocol level, rather than at the redis-py level. This allows for many improvements and bug fixes.

  • #225 Support redis-py 3.0
  • #65 Support execute_command method
  • #206 Drop Python 2.6 support
  • #141 Support strings in integer arguments
  • #218 Watches checks commands rather than final value
  • #220 Better support for calling into redis from Lua
  • #158 Better timestamp handling
  • Support for register_script function.
  • Fixes for race conditions caused by keys expiring mid-command
  • Disallow certain commands in scripts
  • Fix handling of blocking commands inside transactions
  • Fix handling of PING inside pubsub connections

It also has new unit tests based on hypothesis, which has identified many corner cases that are now handled correctly.

1.0rc1

Compared to 1.0b1:

  • #231 Fix setup.py, fakeredis is directory/package now
  • Fix some corner case handling of +0 vs -0
  • Fix pubsub get_message with a timeout
  • Disallow certain commands in scripts
  • Fix handling of blocking commands inside transactions
  • Fix handling of PING inside pubsub connections
  • Make hypothesis tests skip if redis is not running
  • Minor optimisations to zset

1.0b1

Version 1.0 is a major rewrite. It works at the redis protocol level, rather than at the redis-py level. This allows for many improvements and bug fixes.

  • #225 Support redis-py 3.0
  • #65 Support execute_command method
  • #206 Drop Python 2.6 support
  • #141 Support strings in integer arguments
  • #218 Watches checks commands rather than final value
  • #220 Better support for calling into redis from Lua
  • #158 Better timestamp handling
  • Support for register_script function.
  • Fixes for race conditions caused by keys expiring mid-command

It also has new unit tests based on hypothesis, which has identified many corner cases that are now handled correctly.

0.16.0

  • #224 Add __delitem__
  • Restrict to redis<3

0.15.0

  • #219 Add SAVE, BGSAVE and LASTSAVE commands
  • #222 Fix deprecation warnings in Python 3.7

0.14.0

This release greatly improves support for threads: the bulk of commands are now thread-safe, lock has been rewritten to more closely match redis-py, and pubsub now supports run_in_thread:

  • #213 pipeline.watch runs transaction even if no commands are queued
  • #214 Added pubsub.run_in_thread as it is implemented in redis-py
  • #215 Keep pace with redis-py for zrevrange method
  • #216 Update behavior of lock to behave closer to redis lock

0.13.1

  • #208 eval's KEYS and ARGV are now lua tables
  • #209 Redis operation that returns dict now converted to Lua table when called inside eval operation
  • #212 Optimize _scan()

0.13.0.1

  • Fix a typo in the Trove classifiers

0.13.0

  • #202 Function smembers returns deepcopy
  • #205 Implemented hstrlen
  • #207 Test on Python 3.7

0.12.0

  • #197 Mock connection error
  • #195 Align bool/len behaviour of pipeline
  • #199 future.types.newbytes does not encode correctly

0.11.0

  • #194 Support score_cast_func in zset functions
  • #192 Make __getitem__ raise a KeyError for missing keys

0.10.3

This is a minor bug-fix release.

  • #189 Add 'System' to the list of libc equivalents

0.10.2

This is a bug-fix release.

  • #181 Upgrade twine & other packaging dependencies
  • #106 randomkey method is not implemented, but is not in the list of unimplemented commands
  • #170 Prefer readthedocs.io instead of readthedocs.org for doc links
  • #180 zadd with no member-score pairs should fail
  • #145 expire / _expire: accept 'long' also as time
  • #182 Pattern matching does not match redis behaviour
  • #135 Scan includes expired keys
  • #185 flushall() doesn't clean everything
  • #186 Fix psubscribe with handlers
  • Run CI on PyPy
  • Fix coverage measurement

0.10.1

This release merges the fakenewsredis fork back into fakeredis. The version number is chosen to be larger than any fakenewsredis release, so version numbers between the forks are comparable. All the features listed under fakenewsredis version numbers below are thus included in fakeredis for the first time in this release.

Additionally, the following was added: - #169 Fix set-bit

fakenewsredis 0.10.0

  • #14 Add option to create an instance with non-shared data
  • #13 Improve emulation of redis -> Lua returns
  • #12 Update tox.ini: py35/py36 and extras for eval tests
  • #11 Fix typo in private method name

fakenewsredis 0.9.5

This release makes a start on supporting Lua scripting: - #9 Add support for StrictRedis.eval for Lua scripts

fakenewsredis 0.9.4

This is a minor bugfix and optimization release: - #5 Update to match redis-py 2.10.6 - #7 Set with invalid expiry time should not set key - Avoid storing useless expiry times in hashes and sorted sets - Improve the performance of bulk zadd

fakenewsredis 0.9.3

This is a minor bugfix release: - #6 Fix iteration over pubsub list - #3 Preserve expiry time when mutating keys - Fixes to typos and broken links in documentation

fakenewsredis 0.9.2

This is the first release of fakenewsredis, based on fakeredis 0.9.0, with the following features and fixes:

  • fakeredis #78 Behaviour of transaction() does not match redis-py
  • fakeredis #79 Implement redis-py's .lock()
  • fakeredis #90 HINCRBYFLOAT changes hash value type to float
  • fakeredis #101 Should raise an error when attempting to get a key holding a list)
  • fakeredis #146 Pubsub messages and channel names are forced to be ASCII strings on Python 2
  • fakeredis #163 getset does not to_bytes the value
  • fakeredis #165 linsert implementation is incomplete
  • fakeredis #128 Remove _ex_keys mapping
  • fakeredis #139 Fixed all flake8 errors and added flake8 to Travis CI
  • fakeredis #166 Add type checking
  • fakeredis #168 Use repr to encode floats in to_bytes

About

Fake implementation of redis API (redis-py) for testing purposes

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 100.0%