From the profiler (#8108) and the access log, 2026-11-02 04:16Z: stalls are 20 a minute, worst 25 s.
1. /board: 51% of main-thread samples. One signed-in Safari tab made 361 GETs in 15 minutes, every 2.5 s, 1.7 s each. Inside it: returnsStrip 29% (tokenReturns, returns.ts:187, then pillarOfIssue 123,000 calls a minute with the closesFromBody regex), forgeNotes → heldChecks 9.5%, today → plannerHold → readyNow → holdsUpMap 11%.
2. diskBytes (apps/contain.ts:943): synchronous lstat walk, 9.4%.
3. rmSync in apps/runtime.ts:1228 execute: 6.9%.
4. The statement tally (slow_sql.ts:95, callers): 8.1%: a stack per statement.
5. The profiler (perf/sample.ts) stays attached until restart, so a sample leaves the forge slower: stop it at the end of the sample, or only sample in a way that can stop.
Do in this order, separate small pulls, land each at once:
A live page never refetches per event. At most one refetch per 30 s per tab, none while hidden; serve /board from the stale-while-revalidate cache home uses. Check other live pages for the same pattern. Do this first: it is half the problem.
Sample the tally (1 in 50) and make the profiler detach.
tokenReturns: one joined query in place of N+1 pillarOfIssue; store closes-from-body instead of re-parsing it.
diskBytes and rmSync off the main thread.
Done when: stalls over 500 ms are under 5 in 10 minutes with a /board tab open at this load. Paste before and after here. Name each pull Part of this issue, not Closes, until the last.
From the profiler (#8108) and the access log, 2026-11-02 04:16Z: stalls are 20 a minute, worst 25 s.
1. /board: 51% of main-thread samples. One signed-in Safari tab made 361 GETs in 15 minutes, every 2.5 s, 1.7 s each. Inside it:
returnsStrip29% (tokenReturns,returns.ts:187, thenpillarOfIssue123,000 calls a minute with theclosesFromBodyregex),forgeNotes → heldChecks9.5%,today → plannerHold → readyNow → holdsUpMap11%. 2. diskBytes (apps/contain.ts:943): synchronous lstat walk, 9.4%. 3. rmSync inapps/runtime.ts:1228 execute: 6.9%. 4. The statement tally (slow_sql.ts:95,callers): 8.1%: a stack per statement. 5. The profiler (perf/sample.ts) stays attached until restart, so a sample leaves the forge slower: stop it at the end of the sample, or only sample in a way that can stop.Do in this order, separate small pulls, land each at once:
/boardfrom the stale-while-revalidate cache home uses. Check other live pages for the same pattern. Do this first: it is half the problem.tokenReturns: one joined query in place of N+1pillarOfIssue; store closes-from-body instead of re-parsing it.Done when: stalls over 500 ms are under 5 in 10 minutes with a /board tab open at this load. Paste before and after here. Name each pull
Part ofthis issue, notCloses, until the last.