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
{{ message }}
Repository navigation
Load-sensitive tests: bucket_transport::fleet_hammer_distinct_branches_all_land_without_coordinator fails repeatedly under load; fill::tests::a_wide_level_is_chunked_by_count_... runs >20 CPU-minutes locally; readers_and_writers_stay_consistent_while_compaction_runs hung once #271
Reported by several agents' full local runs while many cargo builds/tests ran in parallel (CI passes): (1) bucket_transport::tests::fleet_hammer_distinct_branches_all_land_without_coordinator failed in 3 consecutive full runs under load (and once in earlier runs), passes alone; it exercises the manifest CAS retry budget (16 attempts) — under load a legitimate writer may exhaust its retries, which would be a real robustness bug, not only a test problem. (2) fill::tests::a_wide_level_is_chunked_by_count_before_anything_is_split_by_size ran >20 CPU minutes locally. (3) readers_and_writers_stay_consistent_while_compaction_runs hung >30 min once (being investigated in #264). Needed: root-cause each (livelock vs starvation vs unbounded work), bound the retry/backoff behavior (jittered backoff, fair retries), give worker threads hard deadlines so tests fail with a diagnostic instead of hanging, and make test cost deterministic.
Activity
fix(repo): keep the fleet watchdog alive until every worker joins; dr…
fix(repo): backoff between manifest CAS retries; deterministic load-s…