Add test_fact_state_unchanged_throughout_quorum_loss_outage to test_commit_barrier_reply.c (30->31), no production changes. Drives a real repl_remove_standby quorum-loss window through all 4 fact-receipt handlers, polls live state and committed_lsn at 5 time-spread points (not once), asserts a real hook-spy never sees AFTER_EVENT fire, and confirms durable WAL presence via direct on-disk read. Documents an architectural asymmetry for task 6 planning: facts have no commit-timeout concept (COMMIT_MODE_RETAIN registers no continuation, "never honors the timeout") unlike commands' commit_submit_async -- the apply pump must discover newly-committed retained facts purely via committed_lsn advancement, not any per-fact signal. Closes out task-group 4 (fact apply-after-commit split). |
||
|---|---|---|
| .. | ||
| settings | ||
| specs | ||
| steering | ||