Repository navigation
Count the fast store toward presence when the slow store is read-only - #2959
Merged
MarcusSorealheis merged 5 commits intoOct 19, 2026
Merged
MarcusSorealheis merged 5 commits into
MarcusSorealheis merged 5 commits into
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
corcillo
reviewed
Oct 19, 2026
corcillo
reviewed
Oct 19, 2026
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What and why
FastSlowStore::has_with_resultsanswers from the slow store alone, so a blob only the fast store holds is re-uploaded to the slow one. Withslow_direction: read_only(orget) the slow store never receives uploads: the fast store is the only place a new blob can be, yet presence skipped it. Clients were told to upload blobs they had already stored, and a reader that missed its own fast tier was told the blob did not exist.Across instances this matters when the fast store is shared: on a three-pod CAS with an lz4 compression tier over one backend in front of read-only raw shards, every blob stored after the switch was missing on the other two pods. A per-instance fast tier (memory, local filesystem) still answers only for what that instance stored.
With a read-only or get-only slow store the fast store is now checked first and the slow store is asked about the misses only. A writable slow store keeps the old behaviour.
How was this verified?
Three tests in
fast_slow_store_test.rs: read-only and get-only slow stores find an upload only the fast store holds and an older blob only the slow store holds; a writable slow store still reports a fast-only blob missing. On a bed, fresh uploads to one CAS pod were missing on the other two before the change and present after it.Risk
Low: the new branch runs only when the slow store is read-only or get-only, where nothing is ever re-uploaded to it.
AI assistance
An agent drafted the change and I reviewed every line.