Tests that start a notifier close it, so its pager does not outlive the test and log errors into the next file's counts

grokbot commented

2026-10-17 20:01:20 +00:00

Owner

startNotifier's stop() ends telling but, on purpose, leaves the drain pager
running (main.ts stops the notifier when a drain starts and the drain must
still be paged; notify-drain.test.ts holds that). Only close() clears the
pager. Four test files ended their notifier with stop(), so its pager kept
firing for the life of the bun test process.

test/drill-board-line.test.ts starts one with clockMs: 10: after its afterAll
removes the data directory, the pager logs "notify failed ... ENOENT: statfs
'/forge-drill-line-…'" every 10 ms into the process-wide error ring.
Any later file in the same process that counts errors goes red: vitals'
"errors are counted, bounded, redacted, and a burst is said" (lastHour is
not 1) and tui-sentinel's "an admin has it, and it says what the board says"
(the hunt files an error-rate finding). That is what the forge's parallel
check, which runs many files per process, saw on pulls that never touched
either.

On main, bun test test/drill-board-line.test.ts test/vitals.test.ts test/tui-sentinel.test.ts fails both, every time; with this change it is
25/25 green. accounts, watches and snapshot-mirror had the same stop(); they
are not red today but leak the same interval, so they close too.

Validation

pipeline/validate on 7abaa7fb: pass in 12s — bun 1.3.14, CI-shaped env (scrubbed vars, runner PATH), exact check.yml shard + forge.json commands.

step	result	time
`bun install --frozen-lockfile`	ok (exit 0)	0s
`forge.json check.fast`	ok (exit 0)	3s
`check.yml gate (1) pull_request`	ok (exit 0)	2s
`check.yml gate (2) pull_request`	ok (exit 0)	2s
`check.yml gate (3) pull_request`	ok (exit 0)	5s

⚠️ node v20.19.2 here, v22 on CI (tsc runs under node)

test output (failures, totals)

▶ bun install --frozen-lockfile
  ok  exit 0  0s
▶ forge.json check.fast
Ran 129 tests across 11 files in 2.0s (8 shards).
  ok  exit 0  3s
▶ check.yml gate (1) pull_request
Ran 33 tests across 3 files in 1.7s (3 shards).
  ok  exit 0  2s
▶ check.yml gate (2) pull_request
Ran 20 tests across 4 files in 1.2s (4 shards).
  ok  exit 0  2s
▶ check.yml gate (3) pull_request
Ran 76 tests across 4 files in 5.3s (4 shards).
  ok  exit 0  5s

startNotifier's stop() ends telling but, on purpose, leaves the drain pager running (main.ts stops the notifier when a drain starts and the drain must still be paged; notify-drain.test.ts holds that). Only close() clears the pager. Four test files ended their notifier with stop(), so its pager kept firing for the life of the bun test process. test/drill-board-line.test.ts starts one with clockMs: 10: after its afterAll removes the data directory, the pager logs "notify failed ... ENOENT: statfs '<tmp>/forge-drill-line-…'" every 10 ms into the process-wide error ring. Any later file in the same process that counts errors goes red: vitals' "errors are counted, bounded, redacted, and a burst is said" (lastHour is not 1) and tui-sentinel's "an admin has it, and it says what the board says" (the hunt files an error-rate finding). That is what the forge's parallel check, which runs many files per process, saw on pulls that never touched either. On main, `bun test test/drill-board-line.test.ts test/vitals.test.ts test/tui-sentinel.test.ts` fails both, every time; with this change it is 25/25 green. accounts, watches and snapshot-mirror had the same stop(); they are not red today but leak the same interval, so they close too.  ### Validation `pipeline/validate` on `7abaa7fb`: **pass** in 12s — bun 1.3.14, CI-shaped env (scrubbed vars, runner PATH), exact check.yml shard + forge.json commands. | step | result | time | |---|---|---| | `bun install --frozen-lockfile` | ok (exit 0) | 0s | | `forge.json check.fast` | ok (exit 0) | 3s | | `check.yml gate (1) pull_request` | ok (exit 0) | 2s | | `check.yml gate (2) pull_request` | ok (exit 0) | 2s | | `check.yml gate (3) pull_request` | ok (exit 0) | 5s | - ⚠️ node v20.19.2 here, v22 on CI (tsc runs under node) <details><summary>test output (failures, totals)</summary> ``` ▶ bun install --frozen-lockfile ok exit 0 0s ▶ forge.json check.fast Ran 129 tests across 11 files in 2.0s (8 shards). ok exit 0 3s ▶ check.yml gate (1) pull_request Ran 33 tests across 3 files in 1.7s (3 shards). ok exit 0 2s ▶ check.yml gate (2) pull_request Ran 20 tests across 4 files in 1.2s (4 shards). ok exit 0 2s ▶ check.yml gate (3) pull_request Ran 76 tests across 4 files in 5.3s (4 shards). ok exit 0 5s ``` </details>

grokbot added 1 commit

2026-10-17 20:01:20 +00:00

Tests that start a notifier close it, so its pager does not outlive the test and log errors into the next file's counts

check / gate (2) (pull_request) Successful in 3m28s

Details

check / gate (3) (pull_request) Successful in 4m14s

Details

check / gate (1) (pull_request) Successful in 10m5s

Details

check / check (pull_request) Successful in 1s

Details

preview / preview (pull_request) Has been skipped

Details

7abaa7fb47

startNotifier's stop() ends telling but, on purpose, leaves the drain pager
running (main.ts stops the notifier when a drain starts and the drain must
still be paged; notify-drain.test.ts holds that). Only close() clears the
pager. Four test files ended their notifier with stop(), so its pager kept
firing for the life of the bun test process.

test/drill-board-line.test.ts starts one with clockMs: 10: after its afterAll
removes the data directory, the pager logs "notify failed ... ENOENT: statfs
'<tmp>/forge-drill-line-…'" every 10 ms into the process-wide error ring.
Any later file in the same process that counts errors goes red: vitals'
"errors are counted, bounded, redacted, and a burst is said" (lastHour is
not 1) and tui-sentinel's "an admin has it, and it says what the board says"
(the hunt files an error-rate finding). That is what the forge's parallel
check, which runs many files per process, saw on pulls that never touched
either.

On main, `bun test test/drill-board-line.test.ts test/vitals.test.ts
test/tui-sentinel.test.ts` fails both, every time; with this change it is
25/25 green. accounts, watches and snapshot-mirror had the same stop(); they
are not red today but leak the same interval, so they close too.

grokbot merged commit 351d2db24f into main

2026-10-17 20:57:38 +00:00

grokbot referenced this pull request from a commit

2026-10-17 20:57:40 +00:00

Merge pull request 'Tests that start a notifier close it, so its pager does not outlive the test and log errors into the next file's counts' (#50) from grokbot/forge:fix/tests-close-notifiers into main

Rows
Columns

Tests that start a notifier close it, so its pager does not outlive the test and log errors into the next file's counts #50

Validation