Questions about in-memory caching

[finBitorg]

Feature description

Use case

On raspberrypi writing cache file to sdcards kills the sd card. Please add memory only switch to allow the cache to be stored in memory only and maybe with a max cache size parameter.

for example : SESSION = CachedSession('demo_cache', expire_after=300, mem_only=0)

where mem_only=0 is mem only caching with no size limit; mem_only=1000000 is mem only caching with 1000000 bytes/ 1MB max ram usage for cache; mem_only=None/Not specified then use file based cache as normal.

[JWCook]

Some options are:

• A basic in-memory cache (backend='memory')
• A SQLite in-memory database (backend=SQLiteCache(use_memory=True))
• If you wanted to, using one of the options above, you could manually read/write to storage when you choose (for example only on application startup/shutdown).

Requests-cache currently doesn't have an option to limit the cache to a specific size, but that is a feature I've considered adding.

[finBitorg]

Yes this works; I didn't see it in the docs at first. Thanks for this awesome project !!

[finBitorg]

btw this might be documented somewhere but what's the benefit of SQL lite in memory vs basic ?

[JWCook]

There won't be a noticeable difference for most use cases, but for advanced or specialized usage, SQLite will give you some more features. For example:

• If you wanted to do any lower-level work with the underlying database, you'd have all the features of the python sqlite3 module.
• Some operations might be a bit faster in SQLite due to extra indexes, like clearing expired responses (session.cache.delete(expired=True)).
• For large caches, using backup() to copy an in-memory db to disk (or vice versa) would be faster than copying with session.cache.update().

It's possible that one of those backends might be more memory-efficient than the other in some situations, but I haven't benchmarked that in a long time. I'll leave this issue open as a reminder to do that.

[finBitorg]

Ok last question I promise; I see docs for custom filtering function that returns a boolean, but is possible to specify a custom cache time using a custom function? For example of the response is less than a megabyte cache it for 1hour else if more than ache it for only 20 minutes ?

[JWCook]

There's not an option to provide a callback for setting expiration time, but here's an example that accomplishes the same thing:

from requests_cache import CachedSession
from datetime import datetime, timedelta

session = CachedSession(expire_after=timedelta(hours=1))
r = session.get(url)

# Set a shorter expiration for responses over 1MB
if len(r.content) > 1000000:
    expires = datetime.utcnow() + timedelta(minutes=20)  # Requires an absolute datetime in UTC
    session.cache.save_response(r, expires=expires)