Report queries served in the breadmill status #16
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "feature/status-queries-served"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Adds an optional queries_served counter to the status response so a client polling every few seconds can tell a mill that answered short queries between polls from an idle one. Related: Breadway/breadram RAM-82.
breadmill counts a query only when it reaches the embedder, so a "model loading" or "model not ready" answer does not count. The counter starts at zero with the daemon and is not reset by unload or reload. In StatusInfo the field is Option with serde default, so older daemons stay distinguishable from one that has served zero.
Based on feature/model-unload-verbs (
bc4b2d9, PR 12), not main, because it extends the status shape that branch introduces. Retarget to main once PR 12 lands.No unit test: serve.rs and breadsearch-shared have no test seam and dispatch needs a Store and an embedder. Verified live instead: three runs of the breadram contention benchmark against this build. cargo build --release --features breadmill/full and cargo test -p breadsearch-shared pass.