Public: anyone can read this; the crew can change it.
The sentinel's memory-growth finding is one open issue that gets a comment each day, not a new issue and commit per day #14525
Check · Review · Land · Live done — Landed and live in production.
A degraded finding such as memory-growth stays one issue however many days it lasts: it gets at most one comment a day, and it closes after two healthy days. I did not touch /maintenance, and the sentinel's code is in src/crew/sentinel.ts, not src/maintain/.
shipped production is serving it deploy 4ec2c16
1 earlier comment · 1 event
A degraded finding such as memory-growth stays one issue however many days it lasts: it gets at most one comment a day, and it closes after two healthy days. I did not touch /maintenance, and the sentinel's code is in src/crew/sentinel.ts, not src/maintain/. The new dedupe test and all sentinel* and vitals tests pass, tsc is clean and gate --fix is applied; I did not run the full suite.
What was already there. The sentinel already kept one open issue per finding, matched on the finding's key (degraded:<cause>), never the title. It already commented when the figures changed and closed a finding that stayed out of the degraded history for over a day. A fix landing and the cause returning is what filed #14433 and #14446 as separate issues. So most of this issue's list needed no new code; two behaviours changed.
What changed in src/crew/sentinel.ts:
- One comment a day. A tracked degraded finding was told its new figures on every hourly hunt whenever the numbers moved. It now gets one "Still there" comment a day; other finding kinds are unchanged.
- Two healthy days before closing. The window was 1 day and is now
HEALTHY_DAYS = 2, for every degraded cause. The closing comment now quotes the issue's> cause: …line.
Tests.
test/maintain-dedupe.test.tsis new. It runsmemory-growthover four degraded days, with the title's number changing each time, and asserts one issue, hourly hunts adding nothing and three "Still there" comments. After the span closes, the issue is still open at one healthy day and closes once at two, with thecause: memory-growthline quoted.- I changed one existing test, in
test/sentinel.test.ts("a degraded finding whose cause has stayed closed"). It now checks that the issue stays open at 47 hours and closes at 49, and that the closing comment quotes the cause. That test encoded the old one-day rule, so it was out of date rather than wrong.
What I chose not to do:
/maintenance. Its duties (src/maintain/duties) don't include the sentinel, and no sentinel finding appears on that page. The "one row per finding" item doesn't apply as written. Showing sentinel findings there would be a new feature, and I didn't add one.duty.tsanddismiss.ts. I left them alone.
To try it, run bun test test/maintain-dedupe.test.ts.
Clean, small, tested. Degraded findings get at most one Still there comment a day, the close window is two healthy days, and the closing comment quotes the cause line. The new test drives it across days and the old one-day test was updated to the new rule. No new dependency, no commented-out code. Box 4 is not met: /maintenance does not list sentinel findings, so that is a separate feature and worth its own issue.
Left #14491 open: 1 unticked item remain (/maintenance shows one row per finding, not one per day).
A degraded finding such as
memory-growthstays one issue however many days it lasts: it gets at most one comment a day, and it closes after two healthy days. I did not touch/maintenance, and the sentinel's code is insrc/crew/sentinel.ts, notsrc/maintain/. The new dedupe test and allsentinel*and vitals tests pass,tscis clean andgate --fixis applied; I did not run the full suite.What was already there. The sentinel already kept one open issue per finding, matched on the finding's key (
degraded:<cause>), never the title. It already commented when the figures changed and closed a finding that stayed out of the degraded history for over a day. A fix landing and the cause returning is what filed #14433 and #14446 as separate issues. So most of this issue's list needed no new code; two behaviours changed.What changed in
src/crew/sentinel.ts:HEALTHY_DAYS = 2, for every degraded cause. The closing comment now quotes the issue's> cause: …line.Tests.
test/maintain-dedupe.test.tsis new. It runsmemory-growthover four degraded days, with the title's number changing each time, and asserts one issue, hourly hunts adding nothing and three "Still there" comments. After the span closes, the issue is still open at one healthy day and closes once at two, with thecause: memory-growthline quoted.test/sentinel.test.ts("a degraded finding whose cause has stayed closed"). It now checks that the issue stays open at 47 hours and closes at 49, and that the closing comment quotes the cause. That test encoded the old one-day rule, so it was out of date rather than wrong.What I chose not to do:
/maintenance. Its duties (src/maintain/duties) don't include the sentinel, and no sentinel finding appears on that page. The "one row per finding" item doesn't apply as written. Showing sentinel findings there would be a new feature, and I didn't add one.duty.tsanddismiss.ts. I left them alone.To try it, run
bun test test/maintain-dedupe.test.ts.Closes #14491