- Cache response data for async and non-async path operation functions.
- Lifetime of cached data is configured separately for each API endpoint.
- Requests with
Cache-Control
header containingno-cache
orno-store
are handled correctly (all caching behavior is disabled). - Requests with
If-None-Match
header will receive a response with status304 NOT MODIFIED
ifETag
for requested resource matches header value.
pip install fastapi-redis-cache
Create a FastApiRedisCache
instance when your application starts by defining an event handler for the "startup"
event as shown below:
import os
from fastapi import FastAPI, Request, Response
from fastapi_redis_cache import FastApiRedisCache, cache
from sqlalchemy.orm import Session
LOCAL_REDIS_URL = "redis://127.0.0.1:6379"
app = FastAPI(title="FastAPI Redis Cache Example")
@app.on_event("startup")
def startup():
redis_cache = FastApiRedisCache()
redis_cache.init(
host_url=os.environ.get("REDIS_URL", LOCAL_REDIS_URL),
prefix="myapi-cache",
response_header="X-MyAPI-Cache",
ignore_arg_types=[Request, Response, Session]
)
After creating the instance, you must call the init
method. The only required argument for this method is the URL for the Redis database (host_url
). All other arguments are optional:
host_url
(str
) — Redis database URL. (Required)prefix
(str
) — Prefix to add to every cache key stored in the Redis database. (Optional, defaults toNone
)response_header
(str
) — Name of the custom header field used to identify cache hits/misses. (Optional, defaults toX-FastAPI-Cache
)ignore_arg_types
(List[Type[object]]
) — Cache keys are created (in part) by combining the name and value of each argument used to invoke a path operation function. If any of the arguments have no effect on the response (such as aRequest
orResponse
object), including their type in this list will ignore those arguments when the key is created. (Optional, defaults to[Request, Response]
)- The example shown here includes the
sqlalchemy.orm.Session
type, if your project uses SQLAlchemy as a dependency (as demonstrated in the FastAPI docs), you should includeSession
inignore_arg_types
in order for cache keys to be created correctly (More info).
- The example shown here includes the
Decorating a path function with @cache
enables caching for the endpoint. Response data is only cached for GET
operations, decorating path functions for other HTTP method types will have no effect. If no arguments are provided, responses will be set to expire after one year, which, historically, is the correct way to mark data that "never expires".
# WILL NOT be cached
@app.get("/data_no_cache")
def get_data():
return {"success": True, "message": "this data is not cacheable, for... you know, reasons"}
# Will be cached for one year
@app.get("/immutable_data")
@cache()
async def get_immutable_data():
return {"success": True, "message": "this data can be cached indefinitely"}
Response data for the API endpoint at /immutable_data
will be cached by the Redis server. Log messages are written to standard output whenever a response is added to or retrieved from the cache:
INFO:fastapi_redis_cache:| 04/21/2021 12:26:26 AM | CONNECT_BEGIN: Attempting to connect to Redis server...
INFO:fastapi_redis_cache:| 04/21/2021 12:26:26 AM | CONNECT_SUCCESS: Redis client is connected to server.
INFO:fastapi_redis_cache:| 04/21/2021 12:26:34 AM | KEY_ADDED_TO_CACHE: key=api.get_immutable_data()
INFO: 127.0.0.1:61779 - "GET /immutable_data HTTP/1.1" 200 OK
INFO:fastapi_redis_cache:| 04/21/2021 12:26:45 AM | KEY_FOUND_IN_CACHE: key=api.get_immutable_data()
INFO: 127.0.0.1:61779 - "GET /immutable_data HTTP/1.1" 200 OK
The log messages show two successful (200 OK
) responses to the same request (GET /immutable_data
). The first request executed the get_immutable_data
function and stored the result in Redis under key api.get_immutable_data()
. The second request did not execute the get_immutable_data
function, instead the cached result was retrieved and sent as the response.
In most situations, response data must expire in a much shorter period of time than one year. Using the expire
parameter, You can specify the number of seconds before data is deleted:
# Will be cached for thirty seconds
@app.get("/dynamic_data")
@cache(expire=30)
def get_dynamic_data(request: Request, response: Response):
return {"success": True, "message": "this data should only be cached temporarily"}
NOTE!
expire
can be either anint
value ortimedelta
object. When the TTL is very short (like the example above) this results in a decorator that is expressive and requires minimal effort to parse visually. For durations an hour or longer (e.g.,@cache(expire=86400)
), IMHO, using atimedelta
object is much easier to grok (@cache(expire=timedelta(days=1))
).
Additionally, the decorators listed below define several common durations and can be used in place of the @cache
decorator:
@cache_one_minute
@cache_one_hour
@cache_one_day
@cache_one_week
@cache_one_month
@cache_one_year
For example, instead of @cache(expire=timedelta(days=1))
, you could use:
from fastapi_redis_cache import cache_one_day
@app.get("/cache_one_day")
@cache_one_day()
def partial_cache_one_day(response: Response):
return {"success": True, "message": "this data should be cached for 24 hours"}
If a duration that you would like to use throughout your project is missing from the list, you can easily create your own:
from functools import partial, update_wrapper
from fastapi_redis_cache import cache
ONE_HOUR_IN_SECONDS = 3600
cache_two_hours = partial(cache, expire=ONE_HOUR_IN_SECONDS * 2)
update_wrapper(cache_two_hours, cache)
Then, simply import cache_two_hours
and use it to decorate your API endpoint path functions:
@app.get("/cache_two_hours")
@cache_two_hours()
def partial_cache_two_hours(response: Response):
return {"success": True, "message": "this data should be cached for two hours"}
Below is an example HTTP response for the /dynamic_data
endpoint. The cache-control
, etag
, expires
, and x-fastapi-cache
headers are added because of the @cache
decorator:
$ http "http://127.0.0.1:8000/dynamic_data"
HTTP/1.1 200 OK
cache-control: max-age=29
content-length: 72
content-type: application/json
date: Wed, 21 Apr 2021 07:54:33 GMT
etag: W/-5480454928453453778
expires: Wed, 21 Apr 2021 07:55:03 GMT
server: uvicorn
x-fastapi-cache: Hit
{
"message": "this data should only be cached temporarily",
"success": true
}
- The
x-fastapi-cache
header field indicates that this response was found in the Redis cache (a.k.a. aHit
). The only other possible value for this field isMiss
. - The
expires
field andmax-age
value in thecache-control
field indicate that this response will be considered fresh for 29 seconds. This is expected sinceexpire=30
was specified in the@cache
decorator. - The
etag
field is an identifier that is created by converting the response data to a string and applying a hash function. If a request containing theif-none-match
header is received, theetag
value will be used to determine if the requested resource has been modified.
If this request was made from a web browser, and a request for the same resource was sent before the cached response expires, the browser would automatically serve the cached version and the request would never even be sent to the FastAPI server.
Similarly, if a request is sent with the cache-control
header containing no-cache
or no-store
, all caching behavior will be disabled and the response will be generated and sent as if endpoint had not been decorated with @cache
.
Consider the /get_user
API route defined below. This is the first path function we have seen where the response depends on the value of an argument (user_id: int
). This is a typical CRUD operation where user_id
is used to retrieve a User
record from a database. The API route also includes a dependency that injects a Session
object (db
) into the function, per the instructions from the FastAPI docs:
@app.get("/get_user", response_model=schemas.User)
@cache(expire=3600)
def get_item(user_id: int, db: Session = Depends(get_db)):
return db.query(models.User).filter(models.User.id == user_id).first()
In the Initialize Redis section of this document, the FastApiRedisCache.init
method was called with ignore_arg_types=[Request, Response, Session]
. Why is it necessary to include Session
in this list?
Before we can answer that question, we must understand how a cache key is created. In order to create a unique identifier for the data sent in response to an API request, the following values are combined:
- The optional
prefix
value provided as an argument to theFastApiRedisCache.init
method ("myapi-cache"
). - The module containing the path function (
"api"
). - The name of the path function (
"get_user"
). - The name and value of all arguments to the path function EXCEPT for arguments with a type that exists in
ignore_arg_types
("user_id=1"
).
Therefore, the cache key in this example will be "myapi-cache:api.get_user(user_id=1)"
).
Even though db
is an argument to the path function, it is not included in the cache key because it is a Session
type. If Session
had not been included in the ignore_arg_types
list, caching would be completely broken.
To understand why this is the case, see if you can figure out what is happening in the log messages below:
INFO:uvicorn.error:Application startup complete.
INFO:fastapi_redis_cache.client: 04/23/2021 07:04:12 PM | KEY_ADDED_TO_CACHE: key=myapi-cache:api.get_user(user_id=1,db=<sqlalchemy.orm.session.Session object at 0x11b9fe550>)
INFO: 127.0.0.1:50761 - "GET /get_user?user_id=1 HTTP/1.1" 200 OK
INFO:fastapi_redis_cache.client: 04/23/2021 07:04:15 PM | KEY_ADDED_TO_CACHE: key=myapi-cache:api.get_user(user_id=1,db=<sqlalchemy.orm.session.Session object at 0x11c7f73a0>)
INFO: 127.0.0.1:50761 - "GET /get_user?user_id=1 HTTP/1.1" 200 OK
INFO:fastapi_redis_cache.client: 04/23/2021 07:04:17 PM | KEY_ADDED_TO_CACHE: key=myapi-cache:api.get_user(user_id=1,db=<sqlalchemy.orm.session.Session object at 0x11c7e35e0>)
INFO: 127.0.0.1:50761 - "GET /get_user?user_id=1 HTTP/1.1" 200 OK
The log messages indicate that three requests were received for the same endpoint, with the same arguments (GET /get_user?user_id=1
). However, the cache key that is created is different for each request:
KEY_ADDED_TO_CACHE: key=myapi-cache:api.get_user(user_id=1,db=<sqlalchemy.orm.session.Session object at 0x11b9fe550>
KEY_ADDED_TO_CACHE: key=myapi-cache:api.get_user(user_id=1,db=<sqlalchemy.orm.session.Session object at 0x11c7f73a0>
KEY_ADDED_TO_CACHE: key=myapi-cache:api.get_user(user_id=1,db=<sqlalchemy.orm.session.Session object at 0x11c7e35e0>
The value of each argument is added to the cache key by calling str(arg)
. The db
object includes the memory location when converted to a string, causing the same response data to be cached under three different keys! This is obviously not what we want.
The correct behavior (with Session
included in ignore_arg_types
) is shown below:
INFO:uvicorn.error:Application startup complete.
INFO:fastapi_redis_cache.client: 04/23/2021 07:04:12 PM | KEY_ADDED_TO_CACHE: key=myapi-cache:api.get_user(user_id=1)
INFO: 127.0.0.1:50761 - "GET /get_user?user_id=1 HTTP/1.1" 200 OK
INFO:fastapi_redis_cache.client: 04/23/2021 07:04:12 PM | KEY_FOUND_IN_CACHE: key=myapi-cache:api.get_user(user_id=1)
INFO: 127.0.0.1:50761 - "GET /get_user?user_id=1 HTTP/1.1" 200 OK
INFO:fastapi_redis_cache.client: 04/23/2021 07:04:12 PM | KEY_FOUND_IN_CACHE: key=myapi-cache:api.get_user(user_id=1)
INFO: 127.0.0.1:50761 - "GET /get_user?user_id=1 HTTP/1.1" 200 OK
Now, every request for the same user_id
generates the same key value (myapi-cache:api.get_user(user_id=1)
). As expected, the first request adds the key/value pair to the cache, and each subsequent request retrieves the value from the cache based on the key.
What about this situation? You create a custom dependency for your API that performs input validation, but you can't ignore it because it does have an effect on the response data. There's a simple solution for that, too.
Here is an endpoint from one of my projects:
@router.get("/scoreboard", response_model=ScoreboardSchema)
@cache()
def get_scoreboard_for_date(
game_date: MLBGameDate = Depends(), db: Session = Depends(get_db)
):
return get_scoreboard_data_for_date(db, game_date.date)
The game_date
argument is a MLBGameDate
type. This is a custom type that parses the value from the querystring to a date, and determines if the parsed date is valid by checking if it is within a certain range. The implementation for MLBGameDate
is given below:
class MLBGameDate:
def __init__(
self,
game_date: str = Query(..., description="Date as a string in YYYYMMDD format"),
db: Session = Depends(get_db),
):
try:
parsed_date = parse_date(game_date)
except ValueError as ex:
raise HTTPException(status_code=400, detail=ex.message)
result = Season.is_date_in_season(db, parsed_date)
if result.failure:
raise HTTPException(status_code=400, detail=result.error)
self.date = parsed_date
self.season = convert_season_to_dict(result.value)
def __str__(self):
return self.date.strftime("%Y-%m-%d")
Please note the __str__
method that overrides the default behavior. This way, instead of <MLBGameDate object at 0x11c7e35e0>
, the value will be formatted as, for example, 2019-05-09
. You can use this strategy whenever you have an argument that has en effect on the response data but converting that argument to a string results in a value containing the object's memory location.
If you have any questions, please open an issue. Any suggestions and contributions are absolutely welcome. This is still a very small and young project, I plan on adding a feature roadmap and further documentation in the near future.