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
Fixed — inflated concurrency ("Simultaneous calls") + startup blocked by indexes in schema.sql
Inflated concurrency (~24k vs ~hundreds real): the simultaneous-calls series is a Σstarts − Σends ledger. The rollup's ended query was still slow (70s → cancelled) because the planner preferred the smaller old index idx_calls_ended_ts (no call_result) over v2.126's covering idx_calls_ended_cr → ended was chronically undercounted → the ledger subtracted too little and concurrency grew unbounded. Fix: drop idx_calls_ended_ts (the covering replaces it and forces the index-only scan); the rollup backfill re-populates ended → concurrency rebalances within hours.
Portal startup blocked during the v2.126 update: the covering indexes were in schema.sql, which the updater applies in a BLOCKING way → creating several GB of index over the already-populated calls froze startup ~10 min (capture was never affected; separate service). Fix: large indexes are created ONLY in ensureSearchIndex() with CREATE INDEX CONCURRENTLY (not in schema.sql) → startup no longer blocks on large installs.