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
test: concurrent-invite race test is flaky on the sqlite-file backend (admission_conflict) #855
`modules/tenants/tests/test_owner_race.py::test_concurrent_invites_to_one_address_give_one[sqlite-file]` failed once in CI on PR #854, which does not touch tenants, and passed on re-run:
Five concurrent invites to one address on a SQLite file DB: one loser came back with `admission_conflict` (`ERROR_ADMISSION_CONFLICT`, "someone else changed this organisation at the same moment") instead of `already_invited`. Both refuse the duplicate and only one invitation row exists, so the behaviour is safe; the test asserts the more specific code.
Options:
accept either refusal code in the test for the SQLite backend (the admission conflict is a legitimate outcome under write contention), or
have the invite path re-check for an existing pending invitation after an admission conflict and report `already_invited`.
Activity
test: fix flaky/order-dependent tests (#855 #860 #861) and drop LEGAC…