osd: cache: proxy writes (instead of unconditionally promoting)
This should work similar to the read recency checks that don't always promote on first read, but give the cache osd a choice.
Initially, use a similar (same?) tunable as the reads.
In the future, we can implement other heurstics, like not promoting on write_full (or very large writes), or deciding based on the IO hints in the client request.
- Description updated (diff)
- Target version deleted (
One thing we'll need to be careful about when not promoting is how we handle snapshots. I don't remember exactly how that cooperation works, but I think it won't happen automatically if we aren't fetching the appropriate snaps history from the backend.
- Target version set to v9.0.4
- Status changed from New to In Progress
- Target version changed from v9.0.4 to v9.0.6
- Target version changed from v9.0.6 to v9.0.7
- Target version changed from v9.0.7 to v9.0.8
- Status changed from In Progress to Resolved
Also available in: Atom