Skip to main content

fastapi-cache-stable

Community-maintained fork of fastapi-cache2. The Python import stays fastapi_cache; intentional safety/behavior changes are described in MIGRATION.md.

Latest stable: 0.4.1 on PyPI (released 2026-11-20) · Quickstart · Migration · Compatibility · Security · Support · Changelog · Release provenance. Pre-releases must be independently evaluated before production deployment.

Installation: Python >=3.8,<3.14. CPython 3.11–3.13 (server-side) has security validation; Python 3.8–3.10 is functional legacy compatibility only. DynamoDB/Memcached remain experimental. Browser/Pyodide/Node are untested. Independent project, not endorsed by the upstream maintainer.

stable is the release branch; main tracks upstream.

pypi license CI

Python versions and security

Recommended for production: CPython 3.11–3.13, within our tested FastAPI/Redis/InMemory matrix. Python 3.8–3.10 stays installable and functionally tested to support migrations from fastapi-cache2, but is legacy-only: these EOL runtimes and their dependency stacks receive no security maintenance commitment. Known issues in some Python 3.8/3.9 dependency profiles include AnyIO TLS and Starlette file serving. The library does not exercise those specific paths directly, but other application features might. Use a newer Python and patched compatible dependencies whenever possible.

A vulnerability found in a package shared by this fork and the original fastapi-cache2 may affect both if a consumer resolves the same vulnerable version and invokes the affected code. They do not necessarily install the same dependency versions. In particular, the published fastapi-cache2==0.2.2 and this fork both declared redis-py 4.x in the published 0.3.x metadata. The upstream repository's later, unreleased main branch changed the Redis extra to 5.x. 0.4.x now publishes Python-conditional redis-py ranges through 8.1.x; older 0.3.x artifacts remain unchanged. See compatibility & risk matrix, security policy and legacy vulnerability explanation.

Introduction

fastapi-cache is a tool to cache FastAPI endpoint and function results, with backends supporting Redis, Memcached, and Amazon DynamoDB.

Development roadmap

The published 0.4.x line prioritizes safe, low-friction migration over feature count. See the roadmap and its linked issues for 0.4.x maintenance and 0.5.x work. Features in the roadmap are not included in 0.4.x.

Features

  • Same fastapi_cache imports and legacy default cache-key algorithm.
  • Correctness fixes for cache invalidation, error responses and HTTP cache headers.
  • Auth/Cookie privacy bypass by default; opt-in typed JSON decoding for Pydantic v2.
  • Redis and InMemory backends; Memcached and DynamoDB remain experimental.
  • CI across Python versions, security profiles and published-wheel Redis migration.

Requirements

  • FastAPI with a compatible Python version.
  • Redis standalone verified with Python-conditional redis-py ranges: 4.x through 6.1.x on Python 3.8, through 7.0.x on 3.9, and through 8.1.x on 3.10–3.13.
  • DynamoDB client extra requires Python 3.10+ and patched aiohttp.
  • See COMPATIBILITY.md before using optional backends.

Install

Install the published stable release (pin the version you test):

python -m pip install "fastapi-cache-stable[redis]==0.4.1"
python -m pip check

For process-local InMemory caching, omit [redis]. See the 5-minute quickstart for a complete example.

To evaluate a published pre-release candidate:

python -m pip install --pre --upgrade "fastapi-cache-stable[redis]"

Other extras: [memcache], [dynamodb] (experimental), or no extra for InMemory. Before deploying, check PyPI and pin the exact version you've tested. Never co-install fastapi-cache2; both distributions provide fastapi_cache files.

Migrating from fastapi-cache2

The intended low-friction upgrade is fastapi-cache2==0.2.2 → fastapi-cache-stable, keeping your existing from fastapi_cache ... imports and usual @cache(...) calls. For GET/JSON with standalone Redis or InMemory, the default key algorithm is unchanged. Uninstall the old distribution first: the two packages install the same Python module and must not coexist. Update the dependency in your application's lockfile.

This is not a promise of identical behavior in every case: authenticated requests bypass cache by default, second init() calls now raise, and clear(key=...) is intentionally narrower. Old private/special Redis entries may require targeted invalidation. See the quick migration checklist, verification scenarios and rollback instructions.

Usage

Quick Start

Redis 7 / redis-py 4.x example. Initialize once per application lifespan, then close owned resources and reset the process-global cache configuration on teardown. Avoid routine use of force=True.

from contextlib import asynccontextmanager
from typing import AsyncIterator

from fastapi import FastAPI
from redis import asyncio as aioredis

from fastapi_cache import FastAPICache
from fastapi_cache.backends.redis import RedisBackend
from fastapi_cache.decorator import cache


@asynccontextmanager
async def lifespan(_: FastAPI) -> AsyncIterator[None]:
    redis_client = aioredis.from_url(
        "redis://localhost:6379/0", decode_responses=False
    )
    FastAPICache.init(RedisBackend(redis_client), prefix="fastapi-cache")
    try:
        yield
    finally:
        FastAPICache.reset()
        await redis_client.close()  # redis-py 4.x


app = FastAPI(lifespan=lifespan)


@app.get("/")
@cache(expire=60)
async def index():
    return {"hello": "world"}

For a process-local test fixture, use FastAPICache.init(InMemoryBackend(), prefix="fastapi-cache") after importing InMemoryBackend from fastapi_cache.backends.inmemory. InMemory data is not shared across workers.

Initialization

Call FastAPICache.init() once per process at application startup. A second call raises RuntimeError; test fixtures that re-enter a lifespan must close owned backend resources and call FastAPICache.reset() during teardown. force=True permits deliberate replacement but does not close the old connection pool or isolate multiple FastAPI apps in one process.

Use the @cache decorator

If you want cache a FastAPI response transparently, you can use the @cache decorator between the router decorator and the view function.

Parameter type default description
expire int sets the caching time in seconds
namespace str "" namespace to use to store certain cache items
coder Coder JsonCoder which coder to use, e.g. JsonCoder
key_builder KeyBuilder callable default_key_builder which key builder to use
injected_dependency_namespace str __fastapi_cache prefix for injected dependency keywords.
cache_authenticated bool None per-endpoint override for authenticated requests; requires an identity-aware key_builder.
infer_response_model bool False opt into FastAPI's response-model inference from a return annotation; can enable output validation and break invalid union/Response annotations. Disabled by default for drop-in compatibility.

The cache_status_header argument belongs to FastAPICache.init(), not @cache(). It controls the HIT / MISS response header.

You can also use the @cache decorator on regular functions to cache their result.

Injected Request and Response dependencies

The cache decorator injects dependencies for the Request and Response objects, so that it can add cache control headers to the outgoing response, and return a 304 Not Modified response when the incoming request has a matching If-None-Match header. The ETag is a quoted weak validator derived from a stable SHA-256 digest of the cached representation; If-None-Match supports weak comparison, comma-separated lists and *. Conditional requests are evaluated on both hits and misses, and malformed entity-tags are ignored. This only happens if the decorated endpoint doesn't already list these dependencies already.

The keyword arguments for these extra dependencies are named __fastapi_cache_request and __fastapi_cache_response to minimize collisions. Use the injected_dependency_namespace argument to @cache to change the prefix used if those names would clash anyway.

Supported data types

The default JsonCoder stores any data type that FastAPI can convert to JSON, including Pydantic models and dataclasses. On a cache hit it returns the decoded JSON value (plain dicts/lists), not the annotated model; this is the legacy fastapi-cache2 behavior.

Pydantic v2 users can opt into typed restoration with TypedJsonCoder:

from fastapi_cache import FastAPICache
from fastapi_cache.coder import TypedJsonCoder

FastAPICache.init(backend, coder=TypedJsonCoder)

or per endpoint:

from .models import SomeModel, create_some_model

@app.get("/foo")
@cache(expire=60, coder=TypedJsonCoder)
async def foo() -> SomeModel:
    return create_some_model()

TypedJsonCoder uses the endpoint return annotation to restore the annotated type and propagates validation errors. It requires Pydantic v2; Pydantic v1 and annotations that cannot be represented by a Pydantic v2 schema keep the legacy decoded behavior.

For broader type support, use fastapi_cache.coder.PickleCoder or implement a custom coder (see below).

PickleCoder deserializes with Python's pickle and must only be used with a cache backend whose contents are trusted and integrity-protected. An attacker who can alter cached bytes may be able to execute code when those values are read. JsonCoder is the default.

Optional response-model inference

By default, @cache preserves the upstream behavior of not exposing the function's return annotation as a FastAPI response model. Existing endpoints that return mixed Response and data values therefore keep registering without new Pydantic validation.

To have FastAPI infer an OpenAPI response model from a valid annotation and validate/serialize the output, opt in explicitly:

@app.get("/item")
@cache(expire=60, infer_response_model=True)
async def item() -> SomeModel:
    return create_some_model()

Do not opt in with annotations FastAPI cannot use as a response model, for example Union[Response, dict]. Alternatively, set response_model=SomeModel on @app.get(...) as with ordinary FastAPI. This option changes schema and response validation, not stored cache keys.

Response handling

JsonCoder caches only simple JSONResponse objects: status 200, media type application/json, no custom headers, and no background task. Other Response subclasses are returned directly and are not cached: StreamingResponse, FileResponse, RedirectResponse, plain Response, template responses, and JSONResponse objects with metadata that would be lost on a cache hit. Custom coders can override Coder.can_encode() to opt in; PickleCoder caches response objects it can reproduce faithfully (JSONResponse, plain Response, RedirectResponse) and bypasses streaming, file, and template responses. Responses with Set-Cookie, Cache-Control: no-store or Cache-Control: private, or a background task are never cached, regardless of the coder. If the endpoint or a dependency sets Set-Cookie, Cache-Control: no-store/private, a background task, a non-200 status, or any header other than the automatic content-length on the injected response object, the call is not cached. This covers dict/list returns whose final HTTP metadata comes from that object. User payloads that contain _spec_type are preserved as normal data; the internal marker is namespaced as __fastapi_cache_spec_type__.

Custom coder

By default use JsonCoder, you can write custom coder to encode and decode cache result, just need inherit fastapi_cache.coder.Coder.

from typing import Any
import orjson
from fastapi.encoders import jsonable_encoder
from fastapi_cache import Coder

class ORJsonCoder(Coder):
    @classmethod
    def encode(cls, value: Any) -> bytes:
        return orjson.dumps(
            value,
            default=jsonable_encoder,
            option=orjson.OPT_NON_STR_KEYS | orjson.OPT_SERIALIZE_NUMPY,
        )

    @classmethod
    def decode(cls, value: bytes) -> Any:
        return orjson.loads(value)


@app.get("/")
@cache(expire=60, coder=ORJsonCoder)
async def index():
    return dict(hello="world")

Custom key builder

By default the default_key_builder builtin key builder is used; this creates a cache key from the function module and name, and the positional and keyword arguments converted to their repr() representations, encoded as a MD5 hash. You can provide your own by passing a key builder in to @cache(), or to FastAPICache.init() to apply globally.

For example, if you wanted to use the request method, URL and query string as a cache key instead of the function identifier you could use:

from starlette.requests import Request
from starlette.responses import Response


def request_key_builder(
    func,
    namespace: str = "",
    *,
    request: Request = None,
    response: Response = None,
    args=(),
    kwargs=None,
):
    if request is None:
        raise ValueError("This builder requires an HTTP Request")
    return ":".join([
        namespace,
        request.method.lower(),
        request.url.path,
        repr(sorted(request.query_params.items()))
    ])


@app.get("/")
@cache(expire=60, key_builder=request_key_builder)
async def index():
    return dict(hello="world")

This request-URL builder is appropriate for public data only. A custom builder must include every value that affects the response, particularly validated identity, tenant and permissions when authenticated caching is explicitly enabled. DB session objects with changing repr() values can otherwise produce perpetual misses; see the migration checklist.

Authenticated requests

By default, requests with a non-empty Authorization or Cookie header bypass the cache: the endpoint runs normally and nothing is read from or written to the backend. Public requests keep caching.

This is a conservative default. Session cookies are covered because they use the Cookie header. Empty headers are treated as unauthenticated. Functions decorated without a Request parameter have no HTTP identity context and keep using the cache.

To cache authenticated responses, opt in explicitly and provide a key builder that includes the identity:

import hashlib
import hmac
import os

from starlette.requests import Request

from fastapi_cache import FastAPICache
from fastapi_cache.decorator import cache

CACHE_KEY_SECRET = os.environ["CACHE_KEY_SECRET"].encode()


def identity_key_builder(
    func,
    namespace: str = "",
    *,
    request: Request = None,
    response=None,
    args,
    kwargs,
):
    identity = ""
    if request is not None:
        identity = "|".join(
            [
                request.headers.get("Authorization", ""),
                request.headers.get("Cookie", ""),
                request.headers.get("X-Tenant-Id", ""),
            ]
        )
    key_material = f"{identity}|{args}|{kwargs}"
    digest = hmac.new(
        CACHE_KEY_SECRET,
        key_material.encode(),
        hashlib.sha256,
    ).hexdigest()
    return f"{namespace}:{func.__module__}:{func.__name__}:{digest}"


FastAPICache.init(backend, cache_authenticated=True, key_builder=identity_key_builder)


@app.get("/me")
@cache(expire=60)
async def me(request: Request):
    ...

The key builder must include every dimension that affects the response: token, session, tenant, permissions, and any other validated principal state. Prefer the principal your app already validated over raw headers. Do not put credentials in plaintext in keys or logs; use HMAC-SHA256 with an application secret.

If cache_authenticated=True is used with the default key builder, an authenticated request raises ValueError before the cache is accessed. The flag can also be set per endpoint: @cache(cache_authenticated=True, key_builder=identity_key_builder).

The automatic bypass only detects Authorization and Cookie. If your app uses X-API-Key, proxy-injected headers, request.state, or internal dependencies, review your cache strategy. cache_authenticated=True does not make an unsafe key builder safe.

This protects the application cache. It does not configure CDN, reverse-proxy, or browser caching.

Upgrading from the legacy package or early prereleases: old entries written for authenticated requests may remain under the default key. Clear or rotate the cache namespace before deploying, including all instances/workers and persistent Redis caches. The bypass is not retroactive.

Backend notes

InMemoryBackend

The InMemoryBackend stores cache data in memory and only deletes when an expired key is accessed. This means that if you don't access a function after data has been cached, the data will not be removed automatically.

expire=None means no expiration. Each backend instance owns its store and lock, so a new instance does not see entries from another instance. Do not share one instance across event loops or threads; use one instance per loop/thread. Subclasses that override __init__() must call super().__init__(). InMemory is process-local; use Redis or Memcached for cross-process sharing. expire=0 keeps the legacy fallback to the global expire value. get_with_ttl returns -1 for no-expiration entries; the decorator omits max-age for those entries and for expire=None.

RedisBackend

When using the Redis backend, please make sure you pass in a redis client that does not decode responses (decode_responses must be False, which is the default). Cached data is stored as bytes (binary), decoding these in the Redis client would break caching.

0.4.x publishes Python-conditional Redis client ranges (4.x through 6.1.x on Python 3.8, through 7.0.x on 3.9 and through 8.1.x on 3.10–3.13). See COMPATIBILITY.md and audit/redis/REDIS_CLIENT_MATRIX_RESEARCH.md for real-server evidence. The older 0.3.x artifacts remain immutable.

Tests and coverage

coverage run -m pytest
coverage html
xdg-open htmlcov/index.html

License

This project is licensed under the Apache-2.0 License.

Metadata

Release files for fastapi-cache-stable 0.4.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for fastapi-cache-stable 0.4.1
File Size Uploaded Provenance Actions
fastapi_cache_stable-0.4.1.tar.gz 62.2 kB With provenance Details for fastapi_cache_stable-0.4.1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for fastapi-cache-stable 0.4.1
File Interpreter ABI Platform Actions
fastapi_cache_stable-0.4.1-py3-none-any.whl Python 3 none any Details for fastapi_cache_stable-0.4.1-py3-none-any.whl

Total release size: 90.6 kB

Release files / fastapi_cache_stable-0.4.1.tar.gz

Download URL fastapi_cache_stable-0.4.1.tar.gz
Size 62.2 kB
Tags Source
SHA-256 checksum
How to use checksums
e171db38e12b8a5b1815a5d781018b39d4af883be0d65ed01a6162cfeb0417af
BLAKE2b-256 checksum
How to use checksums
a400c0b7e42c71019e9fc71b0ea1862f4376589e69a328785f71d336acaf5dd1
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 20, 2026.

Transparency log

Release files / fastapi_cache_stable-0.4.1-py3-none-any.whl

Download URL fastapi_cache_stable-0.4.1-py3-none-any.whl
Size 28.4 kB
Tags Python 3
SHA-256 checksum
How to use checksums
a68c770ebdd6dff06e7a6b282b71be1a60dc020b7d3f0b971c307a83489d708d
BLAKE2b-256 checksum
How to use checksums
bf492e6aaf233b049ff2c293c3bb3b9805892669fe1f5c5ebfa07321ca8c153d
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 20, 2026.

Transparency log

Security for fastapi-cache-stable 0.4.1

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for fastapi_cache_stable-0.4.1.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 20, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page