request-cache is not working properly

[meltanolabs]

request-cache is not working properly

[ericboucher]

Idea: store the cache file somewhere else? In S3, or at least somewhere that gives us a bit more control?

https://github.com/marketplace/actions/s3-cache-for-github-actions or similar?

[laurentS]

One year later, I'm reopening this, as the cache is clearly not used.
This screenshot from https://github.com/MeltanoLabs/tap-github/actions/caches confirms why:

Cache is python version specific, which defeats the purpose of our caching attempts.

A local run of the test suite consumes about 500 rest requests and 25 graphql points. As we run 5 versions of python, we're still at only 50% of quota on rest, and much less on graphql. So there's probably some other consumer of quota outside of CI.

Storage of cache outside of github seems like the only option at this point. I will follow up on the suggestion above.

[laurentS]

Actually, I think I misunderstood what is happening. The cache key is api-cache-v4 which is python version agnostic.

But the cached file is 3 weeks old in the screenshot above, so our code ignores it (it expires the cache after 24h) so it hits the API again.

My guess at this point is that the cache works the first day we use it, but it then becomes stale and ignored by requests-cache. 2 options:

• change code to expire the cache after much longer than 24h. The problem being that tests use a start_date of more or less "today". So beyond the first day, it's likely that requests might change
• use a cache key based on the current date, which should make the cache file renew every 24h, in line with the cache expiry and api requests.

[laurentS]

To expand on the above, this is all because cached files are read-only, so the cache file is not updated beyond its first creation time.
See https://github.com/actions/cache/blob/main/tips-and-workarounds.md#update-a-cache for details.