nightly deep gate is red #1757

Activity

1. github-actions commented on Oct 24, 2026

   

   github-actionsBot

   on Oct 24, 2026 – with GitHub Actions

   Author

   More actions

   The nightly deep gate failed.

   Run: https://github.com/commandprompt/pgcolumnar/actions/runs/40086314514

   Jobs that did not succeed:

   • pytest (cluster tests, PG 17)
   • pytest (cluster tests, PG 15)
   • pytest (cluster tests, PG 16)

   This issue is opened by the nightly on its first red, updated on each red
   after it, and closed automatically by the next green run.
2. github-actions commented on Oct 25, 2026

   

   github-actionsBot

   on Oct 25, 2026 – with GitHub Actions

   Author

   More actions

   The nightly deep gate failed.

   Run: https://github.com/commandprompt/pgcolumnar/actions/runs/40186895850

   Jobs that did not succeed:

   • pytest (cluster tests, PG 17)
   • pytest (cluster tests, PG 18)
   • pytest (cluster tests, PG 16)
   • pytest (cluster tests, PG 15)

   This issue is opened by the nightly on its first red, updated on each red
   after it, and closed automatically by the next green run.
3. github-actions commented on Oct 26, 2026

   

   github-actionsBot

   on Oct 26, 2026 – with GitHub Actions

   Author

   More actions

   The nightly deep gate failed.

   Run: https://github.com/commandprompt/pgcolumnar/actions/runs/40322948037

   Jobs that did not succeed:

   • pytest (cluster tests, PG 18)
   • pytest (cluster tests, PG 15)
   • pytest (cluster tests, PG 17)

   This issue is opened by the nightly on its first red, updated on each red
   after it, and closed automatically by the next green run.
4. jdatcmd commented on Oct 26, 2026

   

   jdatcmd

   on Oct 26, 2026

   Collaborator

   More actions

   Cause and fix in #1758. The nightly's cluster-test jobs are cancelled at their 20-minute limit, not failing. On run 40322948037, PG 16 took 18m52s, and PG 15, 17 and 18 hit 20 minutes; PG 15 had already printed 7305 pass + 0 fail. ci.yml got 40 minutes in #1740; nightly.yml did not. The nightly is dispatched on the fix branch (run 40365017512) to prove it.
5. github-actions commented on Oct 26, 2026

   

   github-actionsBot

   on Oct 26, 2026 – with GitHub Actions

   Author

   More actions

   The nightly deep gate failed.

   Run: https://github.com/commandprompt/pgcolumnar/actions/runs/40365017512

   Jobs that did not succeed:

   • coverage report (PG 18)
   • pytest (cluster tests, PG 18)
   • pytest (cluster tests, PG 16)
   • pytest (cluster tests, PG 17)
   • pytest (cluster tests, PG 15)

   This issue is opened by the nightly on its first red, updated on each red
   after it, and closed automatically by the next green run.
6. github-actions commented on Oct 26, 2026

   

   github-actionsBot

   on Oct 26, 2026 – with GitHub Actions

   Author

   More actions

   The nightly deep gate failed.

   Run: https://github.com/commandprompt/pgcolumnar/actions/runs/40367661645

   Jobs that did not succeed:

   • capability name-set sweep (PG 18)

   This issue is opened by the nightly on its first red, updated on each red
   after it, and closed automatically by the next green run.
7. jdatcmd

   added a commit that references this issue on Oct 26, 2026

   Merge pull request #1758 from commandprompt/ci/nightly-cluster-timeout

   ...

   6aec464
8. github-actions commented on Oct 26, 2026

   

   github-actionsBot

   on Oct 26, 2026 – with GitHub Actions

   Author

   More actions

   The nightly deep gate failed.

   Run: https://github.com/commandprompt/pgcolumnar/actions/runs/40376459928

   Jobs that did not succeed:

   • pytest (cluster tests, PG 17)

   This issue is opened by the nightly on its first red, updated on each red
   after it, and closed automatically by the next green run.
9. OffgridwithJD commented on Oct 26, 2026

   

   OffgridwithJD

   on Oct 26, 2026

   Collaborator

   More actions

   @jdatcmd The PG 17 test_a_kill_after_the_stripes_flushed red from run 40376459928 is already fixed in #1760, so no need to duplicate it. The cause was the scenario order:

   • crashA's killed write leaves extended blocks behind.
   • Its metapage reservation is unflushed, so redo restores the old highwater over those blocks.
   • crashB's 100k rows then fit inside them, and the relation never grows.

   The fix pins wal_writer_delay = 10s and runs crashB first. Red-first comes from the pin alone on the old order, and the fix is green 5 of 5 in both suites. Details and measurements are in the PR body.

   🤖 Generated with Claude Code
10. jdatcmd

    closed this as completedin #1760on Oct 26, 2026
11. jdatcmd

    added a commit that references this issue on Oct 26, 2026

    Merge pull request #1760 from commandprompt/test/1444-crash-reservati…

    ...

    5d5d55f