lsm / HyperNeo / 32396946295

80%

push

github

perf(daemon): eliminate full-table COUNT subqueries in sessions.list (#2609)

* perf(daemon): eliminate full-table COUNT subqueries in sessions.list

sessions.list ran two full-table COUNT(*) subqueries (with json_extract per
row) on every debounced re-evaluation, which fires on every sessions-table
change including each visible_message_count bump during message saves. Replace
them with a single-row session_counters aggregate maintained by SQLite
triggers on sessions INSERT/DELETE/UPDATE(type,status,session_context), read
O(1) in mapResult. Migration 197 backfills existing DBs. Measured ~8.2ms ->
~3.0ms per evaluation on a synthetic 5k-session DB (the remaining scan is the
row query, covered by the sibling generated-column task).

* fix(daemon): make session counters robust to corrupt context and self-healing

Address review findings on the sessions.list counter PR:
- Guard json_extract in the counter triggers/backfill with a CASE-based
  json_valid check (SQLite does not short-circuit json_valid OR json_extract;
  only CASE skips the branch), so one corrupt legacy session_context row can
  no longer throw mid-backfill (boot-loop) or make archive/delete fail.
- Export the shared human-session predicate from session-counters.ts and build
  the sessions.list WHERE from it (single source of truth, no drift).
- Run backfillSessionCounters at the end of createTables so a future sessions
  rebuild that drops triggers still self-heals totals on the next boot.
- Tests: corrupt (non-JSON) context in backfill + triggers, and DELETE of
  excluded space/room sessions leaving totals unchanged.

44153 of 59591 branches covered (74.09%)

Branch coverage included in aggregate %.

22 of 22 new or added lines in 6 files covered. (100.0%)

60055 of 71843 relevant lines covered (83.59%)

3860.44 hits per line