You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Found by a memory-accounting audit of 8b3c813. The tests are on branch audit/memory-tests and fail on main today; check out that branch to run them. File and line links point at that commit.
What goes wrong. A disk record is charged 0 memory (src/core/src/cache/cached_batch.rs:65). Its index slot, policy node and t4 index entry still live on the heap, so memory use grows with the disk tier and no budget sees it.
Evidence.
249.4 heap bytes per live disk record in release, or with the debug trace drained. The debug figure of 467 included the event trace (debug builds only).
memory_usage_bytes was 0 throughout.
Projection, not measured: about 2.3 GB per 10 million disk entries.
100,000 entries (AUDIT_N1); without AUDIT_DRAIN=1, a debug build over-reports
The split of the 249 B between the cache's own structures and t4's was not measured.
Flags: P probabilistic (a pass after a fix only means the race wasn't hit), F interleaving forced by a test hook, L Linux only, D needs Docker, M measurement only, C control (passes by design).
Activity