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
Flaky: pass_reads_images_of_due_rows_only_and_reports_earliest_due needs phase 2 within 30 s of phase 1 #990
gsi_worker_pass::pass_reads_images_of_due_rows_only_and_reports_earliest_due in crates/storage-cassandra/tests/direct/gsi_worker_pass.rs fails some CI runs at the test's own setup check:
phase 2: the earliest row of partition 1 is in account 0, not in account 2, which this phase needs
Phase 1 queues 50 rows in account 0 with a 60 s delay and 50 rows in account 1 with a 120 s delay. Phase 2 queues 50 rows in account 2 with a 30 s delay, and expects them to be the earliest rows in every worker partition. That holds only when phase 2 queues its rows less than 30 s after phase 1 queued its rows in account 0. Between the two, the test sends the rest of phase 1's puts, runs two passes of every worker and reads every account's queue. The failure fits that interval running past 30 s on this runner.
Proposed fix: keep the order of the three delays and widen the gaps between them. For example, the delays could be 300 s for account 2, 3,600 s for account 0 and 7,200 s for account 1. The only row that becomes due during the test is the 1 ms row of phase 1. The expected next due times come from the queued rows, so longer delays change nothing else the test checks.
Activity