redipy is a Python library that provides a uniform interface to Redis-like
storage systems. It allows you to use the same Redis API with different backends
that implement the same functionality, such as:
redipy.memory: A backend that runs inside the current process and stores data in memory using Python data structures.redipy.redis: A backend that connects to an actual Redis instance and delegates all operations to it.
Table of Contents
This medium article explores some of the rationale behind the library.
If you need certain functionality or found a bug, have a look at the contributing section.
You can install redipy using pip:
pip install redipyTo use redipy, you need to import the library and create a redipy client
object with the desired backend. For example:
# Import the redipy library
import redipy
# Create a redipy client using the memory backend
r = redipy.Redis()
# Create a redipy client using the redis backend
r = redipy.Redis(host="localhost", port=6379)
# Or preferred
r = redipy.Redis(
cfg={
"host": "localhost",
"port": 6379,
"passwd": "",
# A prefix that gets added to every key.
# Can be used to implement namespaces.
"prefix": "",
})
# You can specify the backend explicitly to ensure that the correct parameters
# are passed to the constructor
r = redipy.Redis(
backend="redis",
cfg={
"host": "localhost",
"port": 6379,
"passwd": "",
"prefix": "",
})The redipy client object supports similar methods and attributes to the
official Redis Python client library.
You can use them as you would normally do with redis. For example:
# Set some values
r.set_value("foo", "bar")
r.set_value("baz", "qux")
# Get some values
r.get_value("foo") # "bar"
r.get_value("baz") # "qux"
# Push some values
r.lpush("mylist", "a", "b", "c")
r.rpush("mylist", "d")
# Pop values
r.lpop("mylist") # "c"
r.rpop("mylist", 3) # ["d", "a", "b"]More examples can be found in the examples folder.
The main features of redipy are:
-
Flexibility: You can choose from different backends that suit your needs and preferences, without changing your code or learning new APIs.
-
Adaptability: You can start your project small with the memory backend and only switch to a full Redis server once the application grows.
-
Scripting: You can create backend independent Redis scripts without using Lua. Scripts are written using a symbolic API in python.
-
Compatibility: You can use any Redis client or tool with any backend.
-
Mockability: You can use redipy in tests that require Redis with the memory backend to easily mock the functionality without actually having to run a Redis server in the background. Also, this avoids issues that might occur when running tests in parallel with an actual Redis server.
-
Performance: You can leverage the high performance of Redis or other backends that offer fast and scalable data storage and retrieval.
-
Migration: You can easily migrate data between different backends, or use multiple backends simultaneously.
Redis scripts can be defined via a symbolic API in python and can be executed by any backend.
Here, we are writing a filter function that drains a Redis list
and puts items into a "left" and a "right" list by comparing each items
numerical value with a given cmp value:
import redipy
# set up script
ctx = redipy.script.FnContext()
# add argument
cmp = ctx.add_arg("cmp")
# add key arguments
inp = redipy.script.RedisList(ctx.add_key("inp"))
left = redipy.script.RedisList(ctx.add_key("left"))
right = redipy.script.RedisList(ctx.add_key("right"))
# add local variable which contains the current value pop'ed from the list
cur = ctx.add_local(inp.lpop())
# we consume "inp" until it is empty
loop = ctx.while_(cur.ne_(None))
# push the value to the list depending on whether it is smaller than `cmp`
b_then, b_else = loop.if_(redipy.script.ToNum(cur).lt_(cmp))
b_then.add(left.rpush(cur))
b_else.add(right.rpush(cur))
# pop next value and store in local variable
loop.add(cur.assign(inp.lpop()))
# the script doesn't return a value
ctx.set_return_value(None)
# make sure to build the script only once and reuse the filter_list function
filter_list = r.register_script(ctx)
r.rpush("mylist", "1", "3", "2", "4")
filter_list(
keys={
"inp": "mylist",
"left": "small",
"right": "big",
},
args={
"cmp": 3,
})
r.lpop("mylist", 4) # []
r.lpop("small", 4) # ["1", "2"]
r.lpop("big", 4) # ["3", "4"]Here, we are implementing and object stack with fall-through lookup. Each frame
in the stack has its own fields. If the user tries to access a field that
doesn't exist in the current stack frame (and they are using get_cascading)
the accessor will recursively go down the stack until a value for the given
field is found (or the end of the stack is reached).
from typing import cast
from redipy import RedisClientAPI
from redipy.script import (
ExecFunction,
FnContext,
JSONType,
RedisHash,
RedisVar,
Strs,
ToIntStr,
ToNum,
)
class RStack:
"""An example class that simulates a key value stack."""
def __init__(self, rt: RedisClientAPI) -> None:
self._rt = rt
self._set_value = self._set_value_script()
self._get_value = self._get_value_script()
self._pop_frame = self._pop_frame_script()
self._get_cascading = self._get_cascading_script()
def key(self, base: str, name: str) -> str:
"""
Compute the key.
Args:
base (str): The base key.
name (str): The name.
Returns:
str: The key associated with the name.
"""
return f"{base}:{name}"
def push_frame(self, base: str) -> None:
"""
Pushes a new stack frame.
Args:
base (str): The base key.
"""
self._rt.incrby(self.key(base, "size"), 1)
def pop_frame(self, base: str) -> dict[str, str]:
"""
Pops the current stack frame and returns its values.
Args:
base (str): The base key.
Returns:
dict[str, str] | None: The content of the stack frame.
"""
res = self._pop_frame(
keys={
"size": self.key(base, "size"),
"frame": self.key(base, "frame"),
},
args={})
if res is None:
return {}
return cast(dict, res)
def set_value(self, base: str, field: str, value: str) -> None:
"""
Set a value in the current stack frame.
Args:
base (str): The base key.
field (str): The field.
value (str): The value.
"""
self._set_value(
keys={
"size": self.key(base, "size"),
"frame": self.key(base, "frame"),
},
args={"field": field, "value": value})
def get_value(self, base: str, field: str) -> JSONType:
"""
Returns a value from the current stack frame.
Args:
base (str): The base key.
field (str): The field.
Returns:
JSONType: The value.
"""
return self._get_value(
keys={
"size": self.key(base, "size"),
"frame": self.key(base, "frame"),
},
args={"field": field})
def get_cascading(self, base: str, field: str) -> JSONType:
"""
Returns a value from the stack. If the value is not in the current
stack frame the value is recursively retrieved from the previous
stack frames.
Args:
base (str): The base key.
field (str): The field.
Returns:
JSONType: The value.
"""
return self._get_cascading(
keys={
"size": self.key(base, "size"),
"frame": self.key(base, "frame"),
},
args={"field": field})
def _set_value_script(self) -> ExecFunction:
ctx = FnContext()
rsize = RedisVar(ctx.add_key("size"))
rframe = RedisHash(Strs(
ctx.add_key("frame"),
":",
ToIntStr(rsize.get_value(default=0))))
field = ctx.add_arg("field")
value = ctx.add_arg("value")
ctx.add(rframe.hset({
field: value,
}))
ctx.set_return_value(None)
return self._rt.register_script(ctx)
def _get_value_script(self) -> ExecFunction:
ctx = FnContext()
rsize = RedisVar(ctx.add_key("size"))
rframe = RedisHash(Strs(
ctx.add_key("frame"),
":",
ToIntStr(rsize.get_value(default=0))))
field = ctx.add_arg("field")
ctx.set_return_value(rframe.hget(field))
return self._rt.register_script(ctx)
def _pop_frame_script(self) -> ExecFunction:
ctx = FnContext()
rsize = RedisVar(ctx.add_key("size"))
rframe = RedisHash(Strs(
ctx.add_key("frame"),
":",
ToIntStr(rsize.get_value(default=0))))
lcl = ctx.add_local(rframe.hgetall())
ctx.add(rframe.delete())
b_then, b_else = ctx.if_(ToNum(rsize.get_value(default=0)).gt_(0))
b_then.add(rsize.incrby(-1))
b_else.add(rsize.delete())
ctx.set_return_value(lcl)
return self._rt.register_script(ctx)
def _get_cascading_script(self) -> ExecFunction:
ctx = FnContext()
rsize = RedisVar(ctx.add_key("size"))
base = ctx.add_local(ctx.add_key("frame"))
field = ctx.add_arg("field")
pos = ctx.add_local(ToNum(rsize.get_value(default=0)))
res = ctx.add_local(None)
cur = ctx.add_local(None)
rframe = RedisHash(cur)
loop = ctx.while_(res.eq_(None).and_(pos.ge_(0)))
loop.add(cur.assign(Strs(base, ":", ToIntStr(pos))))
loop.add(res.assign(rframe.hget(field)))
loop.add(pos.assign(pos - 1))
ctx.set_return_value(res)
return self._rt.register_script(ctx)The current limitations of redipy are:
- Some Redis commands are not supported yet: This is likely due to redundant functionality. For all other cases it will eventually be resolved. Check this issue to see the status of redis functions.
- The API differs slightly: Most notably stored values are always strings (i.e., the bytes returned by Redis are decoded as utf-8).
- The semantic of Redis functions inside scripts has been altered to feel more
natural coming from python: Redis functions inside Lua scripts often differ
greatly from the documented behavior. For example,
LPOPreturnsfalsefor an empty list inside Lua (instead ofnilorcjson.null). WhileLPOPreturnsNonein the python API. The script API ofredipyhas been altered to match the python API more closely. As the user doesn't code in Lua directly the benefit of having a more consistent API outweighs the more complicated Lua code that needs to be generated in the backend. - Scripts aim to use python semantics as best as possible: In Lua array indices start at 1. The script API uses a 0 based indexing system and transparently adjusts indices in the Lua backend. Other, similar changes are performed as well.
- Scripts use JSON to pass arguments and return values: The arguments to the
script are passed as JSON bytes for the Lua backend. Keys are passed as is.
The return value of the script is also converted into JSON when moving from
Lua to python. Note, that the empty dictionary (
{}) and the empty list ([]) are indistinguishable in Lua soNoneis returned instead of setting the return value to either of these.
Any contribution, even if it is just creating an issue for a bug, is much appreciated.
If you encounter a bug, please open an issue to draw attention to it or give a thumbsup if the issue already exists. This helps with prioritizing implementation efforts. Even if you cannot solve the bug yourself, investigating why it happens or creating a PR to add test cases helps a lot. If you have a fix for a bug don't hesistate to open a PR.
If you encounter a missing Redis or Lua function please consider adding it yourself (see the implementing section). Here also opening an issue or giving a thumbsup to existing issues helps with prioritization.
However, if you need it only in your local setup
without API support or support for multiple backends, pipelines, etc. you can
use the raw underlying Redis connection via
redipy.main.Redis.get_redis_runtime and
redipy.redis.conn.RedisConnection.get_connection or make use of
the plug-in mechanism.
For the memory backend you can use
redipy.memory.rt.LocalRuntime.add_redis_function_plugin or
redipy.memory.rt.LocalRuntime.add_general_function_plugin. The methods need
a module that contains subclasses of redipy.plugin.LocalRedisFunction and
redipy.plugin.LocalGeneralFunction respectively. Once the new functions are
defined via loading the plugin they can be used in a redipy.script.FnContext
via redipy.script.RedisFn or redipy.script.CallFn respectively.
Note, that redipy.script.RedisFn and redipy.script.CallFn can always be
used in Redis backend scripts. However, calling functions this way will have
the native Lua behavior which can lead to surprising results. To patch those
up as well you can use redipy.redis.lua.LuaBackend.add_redis_patch_plugin,
redipy.redis.lua.LuaBackend.add_general_patch_plugin, and
redipy.redis.lua.LuaBackend.add_helper_function_plugin to add the subclasses
of redipy.plugin.LuaRedisPatch, redipy.plugin.LuaRedisPatch, and
redipy.plugin.HelperFunction respectively. Those functions then can also be
used with the redipy.script.RedisFn and redipy.script.CallFn commands.
Adding functions as described above is discouraged as it may lead to inconsistent support of different backends and inconsistent behavior across different backends.
The easiest way to contribute to redipy is to pick some Redis API functions
that have not (or not completely) been implemented in redipy
yet. It is also much appreciated if you just add test cases or the stubs in a
PR. For a full implementation follow these steps:
- Add the signature of the function to
redipy.api.RedisAPI. Adjust as necessary from the Redis spec to get a pythonic feel. Also, add the signature toredipy.api.PipelineAPIbut withNoneas return value. Additionally, add the redirect to the backend inredipy.main.Redis. - Implement the function in
redipy.redis.conn.RedisConnectionandredipy.redis.conn.PipelineConnection. This should be straightforward as there are not too many changes expected. Don't forget to convert bytes into strings via...decode("utf-8")(there are various helper functions for this inredipy.util). - Add tests to
test/test_sanity.pyto determine the function's behavior in Lua (especially its edge cases). - If the Lua behavior needs to be changed to provide a better feel you can add
a monkeypatch for the function call by either creating a class in
redipy.redis.rpatchto directly change the returned expr for the execution graph or using a Lua helper function via adding a class toredipy.redis.helpers(you need to use a patch to use the helper in the right location). - Next, add and implement the functionality in
redipy.memory.state.Machineand add the appropriate redirects inredipy.memory.rt.LocalRuntimeandredipy.memory.rt.LocalPipeline. - To make the new function accessible in scripts from the memory backend add
a class in
redipy.memory.rfun. - Add the approriate class or method in the right
redipy.symbolic.r...pyfile. If it is a new class / file add an import toredipy.script. - Add a new test in
test/test_api.pyto verify the new function works inside a script for all backends. You can runmake pytest FILE=test/test_api.pyto execute the test andmake coverage-reportto verify that the new code is executed. - Make sure
make lint-allpasses, as well as, all tests (make pytest) run without issue.
You can submit your patch as pull request here.
The changelog can be found here.
redipy is licensed under the Apache License (Version 2.0).
If you have any questions, suggestions, or issues with redipy, please feel
free to open an issue on GitHub. I would love to hear your feedback
and improve redipy. Thank you!