kayg/calternal-v0

Archived

Watch

Fork 0

Merge35b soak reaches 20 overlay cycles before its runner deadline #1490

Open

opened 2026-10-08 23:21:15 +00:00 by kayg · 4 comments

kayg commented

2026-10-08 23:21:15 +00:00

Owner

Merge35b (#1479), head 16250888d. The requested browser soak completed its 30-minute navigation phase and logged 20 of 100 overlay/preview/Composer cycles. The 2,250-second runner deadline then stopped the browser while cycleOverlays waited for the Quick Look level-2 heading at apps/web/e2e/browser-soak.mjs:126. Exit 124 is a runner timeout, not proof of a server crash. No overall soak pass or heap-growth result is claimed. Classification against production is pending. Evidence: artifacts/merge35b/e2e/browser-soak-final.log. The original cycle count and assertions were not changed.

Merge35b (#1479), head 16250888d. The requested browser soak completed its 30-minute navigation phase and logged 20 of 100 overlay/preview/Composer cycles. The 2,250-second runner deadline then stopped the browser while cycleOverlays waited for the Quick Look level-2 heading at apps/web/e2e/browser-soak.mjs:126. Exit 124 is a runner timeout, not proof of a server crash. No overall soak pass or heap-growth result is claimed. Classification against production is pending. Evidence: artifacts/merge35b/e2e/browser-soak-final.log. The original cycle count and assertions were not changed.

kayg referenced this issue

2026-10-08 23:23:41 +00:00

Merge round 35b #1479

kayg commented

2026-10-09 00:55:33 +00:00

Author

Owner

The first full production comparison on cc1af3c12 reproduces the same deadline symptom: exit 124 at the unchanged 2,250-second runner limit, after the 30-minute navigation phase, while waiting for Quick Look's level-2 heading. Last heading wait: cycle 26; last visible heading: cycle 25; last complete ten-cycle batch: 20. Cycle 20's heading is visible and the run continues beyond it. No overall 100-cycle or heap-growth pass is claimed.

This is production evidence for the runner-budget symptom, not evidence of a new server crash. The remaining production and merged runs retain the full original conditions and are pending. Logs: artifacts/m35btriage/soak-prod-1.log and ui-status.txt. Opt-in trace logging changes no assertion or deadline (#1479).

The first full production comparison on cc1af3c12 reproduces the same deadline symptom: exit 124 at the unchanged 2,250-second runner limit, after the 30-minute navigation phase, while waiting for Quick Look's level-2 heading. Last heading wait: cycle 26; last visible heading: cycle 25; last complete ten-cycle batch: 20. Cycle 20's heading is visible and the run continues beyond it. No overall 100-cycle or heap-growth pass is claimed. This is production evidence for the runner-budget symptom, not evidence of a new server crash. The remaining production and merged runs retain the full original conditions and are pending. Logs: artifacts/m35btriage/soak-prod-1.log and ui-status.txt. Opt-in trace logging changes no assertion or deadline (#1479).

kayg commented

2026-10-09 01:39:51 +00:00

Author

Owner

This pair gives no merged-only stall at cycle 20. These are local shared-host deadline observations, not controlled GPU throughput measurements. The second pair remains scheduled. Evidence: artifacts/m35btriage/soak-{prod,merged}-1.log and comparison-summary.json. All original assertions and deadlines remain unchanged.

The first full pair reproduces the original runner-budget finding on both releases. Both run the 30-minute navigation phase, target all 100 cycles and retain the 2,250-second deadline. Production exits 124 with the last complete ten-cycle batch at 20 (last heading visible 25, next heading wait 26). Merged exits 124 with the last complete batch at 40; its interrupted wait is for the Quick Look close button, not the original cycle-20 heading. Both runs show cycle 20's heading and continue beyond it. No completed 100-cycle or final heap-growth result is claimed. This pair gives no merged-only stall at cycle 20. These are local shared-host deadline observations, not controlled GPU throughput measurements. The second pair remains scheduled. Evidence: artifacts/m35btriage/soak-{prod,merged}-1.log and comparison-summary.json. All original assertions and deadlines remain unchanged.

kayg commented

2026-10-09 02:24:36 +00:00

Author

Owner

This is a pre-existing runner-budget finding on the local shared host, not proof of a new crash or a controlled GPU performance regression. Evidence: artifacts/m35btriage/soak-prod-2.log and comparison-summary.json. Keep this issue open.

Production now reproduces the original full-soak runner-budget symptom in 2/2 runs. Both retain 30-minute navigation, 100 cycles, and the 2,250-second deadline. The second run exits 124 in a Quick Look heading wait: last heading wait 62, last visible heading 61, last complete batch 60. Both production runs progress beyond cycle 20. Merged run 1 also exits at the same deadline, after batch 40. Merged run 2 is in progress. No overall 100-cycle or final heap-growth pass is claimed. No assertion or deadline changed. This is a pre-existing runner-budget finding on the local shared host, not proof of a new crash or a controlled GPU performance regression. Evidence: artifacts/m35btriage/soak-prod-2.log and comparison-summary.json. Keep this issue open.

kayg commented

2026-10-09 03:02:13 +00:00

Author

Owner

#1479 final comparison: two original full runs per release (30-minute navigation, 100-cycle target, unchanged 2,250-second runner budget). Both production cc1af3c12 and merged 16250888d hit the runner deadline in 2/2 runs. All four show the cycle-20 Quick Look heading and continue beyond it. Production final visible headings: 25 and 61; merged: 45 and 66. Final complete batches: production 20/60, merged 40/60. Production reproduces the deadline symptom, so this finding is pre-existing and does not block merge readiness. No 100-cycle or final heap-growth pass is claimed; these shared-host local timings are not a controlled GPU regression measurement. Assertions and deadlines remain unchanged. Logs: artifacts/m35btriage/soak-{prod,merged}-{1,2}.log. Keep this issue open.

#1479 final comparison: two original full runs per release (30-minute navigation, 100-cycle target, unchanged 2,250-second runner budget). Both production cc1af3c12 and merged 16250888d hit the runner deadline in 2/2 runs. All four show the cycle-20 Quick Look heading and continue beyond it. Production final visible headings: 25 and 61; merged: 45 and 66. Final complete batches: production 20/60, merged 40/60. Production reproduces the deadline symptom, so this finding is pre-existing and does not block merge readiness. No 100-cycle or final heap-growth pass is claimed; these shared-host local timings are not a controlled GPU regression measurement. Assertions and deadlines remain unchanged. Logs: artifacts/m35btriage/soak-{prod,merged}-{1,2}.log. Keep this issue open.

kayg referenced this issue

2026-10-09 03:04:26 +00:00

Merge round 35b #1479

Commenting is not possible because the repository is archived.

No labels

No milestone

No project

No assignees

1 participant

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

kayg/calternal-v0#1490

No description provided.

Rows
Columns