Skip to content

CI blocker: Chocolatey HTTP 504 during Windows PostgreSQL installation #46

Closed
Closed
CI blocker: Chocolatey HTTP 504 during Windows PostgreSQL installation#46

Description

@ShutenOishi
ShutenOishi
opened on Oct 9, 2026

2026-10-10 JST CI incident: Chocolatey feed 504 during Step 23 rollout

During Step 23's eight extension-local opt-in PR CI runs, multiple Windows PG-major jobs failed before extension compilation in the unchanged shared scripts/install-postgresql.ps1, because the Chocolatey community v2 feed returned HTTP 504 Gateway Timeout when resolving pinned PostgreSQL packages.

Evidence:

These PRs only change config/candidate-build.json. Failures originate from an external pinned-package download/metadata lookup, not from a compiler or functional-test regression. Do not merge any affected PR on failing CI or reduce required checks. Central Step 23 PR build #45 is intentionally blocked until all eight local PRs have green CI and are merged.

Next safe actions

  1. Re-run only failed CI jobs after their runs complete, with the same SHA-pinned build and package versions. Capture pass/fail by PG major; treat repeated 504 as an incident, not as acceptance.
  2. Investigate a bounded retry/backoff strategy for transient 502/503/504 in shared install-postgresql.ps1, with deterministic checksum / version validation and fail-closed behavior. Verify compatibility with the pinned reusable workflow versions in nine caller repos.
  3. Do not fallback to unpinned versions, unverified mirrors, disable attestation, bypass branch protection, or relax required CI.
  4. Accept Step 23 only after CI success and post-merge 9-extension dry-run + candidate-decision artifact audit.

Activity

  1. ShutenOishi commented on Oct 11, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 11, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    2026-10-11 retry pinning in 13 staged manual-only PRs

    The new 14-extension manual-only rollout produced genuine transient Chocolatey failures in older pinned reusable Windows build workflows. Representative new failures occurred before build/functional test in pg_bigm run 38137973709 PG16, pg_cron run 38137983135 PG14 and PG18, pgaudit run 38137979672 PG15, set_user run 38138048203 PG17. PG_bigm PG16 log reports exact pinned postgresql16 16.15.5 Chocolatey Community feed HTTP 504 and exits before compilation.

    Root cause: those existing local Windows workflows pin old Build commit 21c99b933da98f397f739971e738ecb685fe55d2 (or Wave2/3 26be1eed57d8945e5e0a970d432074c530679b8f / 43993b864e59bd35870e15c410ac13c85def8531) whose install-postgresql.ps1 has single-attempt Chocolatey installation. The verified new Build main 455f0ccc667697f600700a6e64bcfd6c57b6a977 contains the existing bounded 3-attempt retry with 10/20-second cooldown, still checks exact Chocolatey version/PG server headers/import library and fails after retries. No unpinned package, mirror, installer bypass or security relaxation.

    All 13 currently UNMERGED extension rollout PR branches were updated to SHA-pin read-only build-extension.yml@455f0ccc667697f600700a6e64bcfd6c57b6a977. Original eight also updated candidate verifier to verify-candidate.yml@ecf84241c1c4cd99fab3f964cb309f9c6e2a01f8 (403 Issues commenting warning handled as report-only). This triggers new exact-head full PG14–18 CI, currently pending/queued, not acceptance. Formal Release builds keep the separately proven SHA-pinned attested reusable workflow @5756b3c5136a8bf4db710016cfe97555f0a71c6d.

    The old failing runs are not deemed success. Keep Issue #46 OPEN until new latest-head 13 PR Windows CI passes and the transient 504 rate is reduced; only then decide closure.

  2. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    Windows CI minimal EDB components — staged PR, NOT deployed (2026-10-15)

    Read-only audit: build/main ff558c6fd967542e78e3eacfc14bbe67326b9223 protected; existing open build PRs #107/#106/#9 unrelated. Central scripts/install-postgresql.ps1 has 3-attempt bounded Chocolatey retry and fail-closed exit/header/library checks. EDB official installer docs permit --disable-components pgAdmin,stackbuilder; Chocolatey documents --ia native argument forwarding. New build PR #110, latest head d6f65be57e72a24cf6d167763a5411396861ea0e: single installer invocation change plus offline PG14–18 argument/retry contract test wired into .github/workflows/validate.yml. No live Windows installation or runtime acceleration measured; PR CI not yet confirmed. Do not merge until exact-head required App15368 CI success.

    Dependency: B2 13 extension update-watch PRs remain independent and must retain existing immutable build pin; central main update alone will NOT alter their CI. After build PR accepted, separately prepare plpgsql_check-only pilot pin update, validate PG14–18 functional builds and ZIP/SHA256/SBOM/Grype and compare installer timings; only then plan 13-repo rollout. #46 HTTP504 retry unchanged. No workflow dispatch, signing canary, Publisher, Release, tag, billing, Secret, permission or branch protection changes.

  3. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    PR #110 exact-head CI recheck — BLOCKED_PENDING (2026-10-15)

    Protected build/main ff558c6fd967542e78e3eacfc14bbe67326b9223; PR #110 latest head d6f65be57e72a24cf6d167763a5411396861ea0e, mergeable=true but mergeable_state=blocked. Exact-head App15368 required resolver is QUEUED in https://github.com/pgextwin/build/actions/runs/38743575514; optional read-only Release audit SKIPPED. No tests have completed in this run. No merge or pilot pin update; existing B2 PRs untouched. Next: wait for resolver SUCCESS; recheck latest head and branch protection, then expected-head squash and postmerge. No Dispatch, signing, Publisher, Release, tag or settings changes.

  4. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    PR #110 recheck: required CI still QUEUED — NO MERGE (2026-10-15)

    Run https://github.com/pgextwin/build/actions/runs/38743575514, attempt 1, exact head d6f65be57e72a24cf6d167763a5411396861ea0e is QUEUED; App15368 required resolver QUEUED, runner_name empty, no steps started or failed. Optional read-only Release job SKIPPED. PR #110 OPEN, mergeable=true but mergeable_state=blocked. Protected build/main unchanged at ff558c6fd967542e78e3eacfc14bbe67326b9223. Fail-closed STOP: no merge, no postmerge run, no plpgsql_check pin pilot, no B2 modifications. This is pending runner scheduling, not evidence of a test failure. Do not rerun/dispatch under current authorization. Recheck exact-head required CI after runner starts.

  5. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    PR #110 repeat audit: still QUEUED; earlier test regression diagnosed (2026-10-15)

    Current PR head d6f65be57e72a24cf6d167763a5411396861ea0e matches Run https://github.com/pgextwin/build/actions/runs/38743575514 (attempt 1), still QUEUED; required App15368 resolver QUEUED; job requests ubuntu-latest, runner_name empty. Protected main ff558c6fd967542e78e3eacfc14bbe67326b9223 unchanged, PR mergeable_state=blocked. Prior run https://github.com/pgextwin/build/actions/runs/38743541404 failed tests/test-postgresql-minimal-install.py at Python regex r"\\d+\\.\\d+\\.\\d+"; current head fixes regex to r"\d+\.\d+\.\d+", but has not been validated by required CI. Repository had successful other CI runs earlier, so persistent general runner outage is NOT established; current job scheduling cause unconfirmed. Runner inventory/Actions permissions endpoints unavailable through read-only connector. STOP: no merge, no pilot SHA update, no B2 changes, no dispatch, no security/publishing changes.

  6. ShutenOishi
    added a commit that references this issue on Oct 15, 2026

    ci: omit pgAdmin and StackBuilder from Windows PostgreSQL installer (#46

    4adc710

  7. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    PR #110 merged after exact-head App15368 PASS (2026-10-15)

    PR #110 exact head d6f65be57e72a24cf6d167763a5411396861ea0e App15368 resolver SUCCESS in https://github.com/pgextwin/build/actions/runs/38743575514; protected main and mergeable_state=clean checked. Expected-head squash MERGED; build/main now 4adc71076283dc9cf64543b760c65b054f5a85f0, protection true. Postmerge push CI https://github.com/pgextwin/build/actions/runs/38744741765 is QUEUED; postmerge NOT YET ACCEPTED. plpgsql_check .github/workflows/windows.yml still pins build reusable workflow at 21c99b933da98f397f739971e738ecb685fe55d2; pilot pin change withheld until postmerge SUCCESS. No B2 edits, no manual dispatch, Publisher, Release, tag, signing or settings changes.

  8. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    #46 minimal EDB install: postmerge gate still pending (2026-10-15)

    Protected build/main 4adc71076283dc9cf64543b760c65b054f5a85f0 unchanged. Postmerge push https://github.com/pgextwin/build/actions/runs/38744741765 is QUEUED; App15368 required resolver QUEUED (ubuntu-latest, no runner assigned), optional signed-public-release read-only job SKIPPED. plpgsql_check protected main 45fcae2a4edcf74170c995a64b6382562e3292cc; its windows.yml remains pinned to 21c99b933da98f397f739971e738ecb685fe55d2. Fail-closed STOP: do not create pilot pin PR before central postmerge SUCCESS. No PG14–18 live installer/functional/ZIP/SHA256/SBOM/Grype/timing claims. No B2 PR changes, Dispatch, signing, Publisher, Release, tags or settings changes.

  9. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    #46 EDB minimal installer Pilot launched (2026-10-15)

    Central protected build/main 4adc71076283dc9cf64543b760c65b054f5a85f0 postmerge https://github.com/pgextwin/build/actions/runs/38744741765 SUCCESS, App15368 resolver SUCCESS. Created plpgsql_check-only PR pgextwin/plpgsql_check#43, head 8d8b1452e7dcb61140040f96498aefae41e82536, exactly one file/one line: windows.yml reusable workflow immutable pin 21c99b933da98f397f739971e738ecb685fe55d2 → 4adc71076283dc9cf64543b760c65b054f5a85f0. At first poll syntax check QUEUED; no PG14–18 Windows acceptance yet. Do not merge before exact-head required CI, five-major smoke, ZIP SHA256 SBOM Grype and timing verification. Existing 13 B2 PRs untouched. No manual Dispatch, signing, Publisher, Release, tags, billing, secrets or permissions changes.

  10. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    plpgsql_check minimal EDB pilot #43 — 5 Windows jobs running, NO MERGE (2026-10-15)

    PR pgextwin/plpgsql_check#43 exact head 8d8b1452e7dcb61140040f96498aefae41e82536; protected main 45fcae2a4edcf74170c995a64b6382562e3292cc unchanged. CI https://github.com/pgextwin/plpgsql_check/actions/runs/38746304977 in_progress. App15368 syntax SUCCESS, matrix resolver SUCCESS, immutable Release preflight SUCCESS; PG14/15/16/17/18 Windows x64 all in_progress, each currently at Install PostgreSQL <major>, no failed steps yet. Required aggregate summary not yet completed; ZIP/SHA256/SBOM/Grype, functional smoke, and per-major install timing NOT VERIFIED. PR mergeable_state=blocked. Fail-closed hold: no merge until exact-head required Windows summary and full five-major artifacts/security/timing acceptance. B2 13 PRs untouched; no dispatch, signature, Publisher, Release, tag or settings changes.

  11. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    plpgsql_check PR #43 live Windows acceptance checkpoint (2026-10-15)

    Exact head 8d8b1452e7dcb61140040f96498aefae41e82536, protected main 45fcae2a4edcf74170c995a64b6382562e3292cc, mergeable_state=blocked. Run https://github.com/pgextwin/plpgsql_check/actions/runs/38746304977 IN_PROGRESS. PG14/15/16/18 jobs SUCCESS; functional smoke, package, SPDX SBOM generate+validate, Grype scan, artifact upload steps SUCCESS. Four artifacts visible (PG14/15/16/18). PG17 install, functional smoke, package, SBOM generate+validate SUCCESS; Grype scan currently completed and upload in_progress, artifact not yet listed. Install step elapsed (GitHub step timestamps) vs old two-run means: PG14 137s vs 532s (-74.2%); PG15 154s vs 557s (-72.4%); PG16 185s vs 536s (-65.5%); PG17 808s vs 137s (+490%); PG18 155s vs 536s (-71.1%). Pilot is single run; PG17 regression is unresolved. SHA256 values and ZIP contents not independently audited (successful packaging metadata steps are not equivalent to independent file inspection). Job log fetch for running PG17 returned 404 BlobNotFound. NO MERGE until all five required checks complete and PG17 slowdown plus artifact integrity are assessed. No B2 edits or forbidden actions.

  12. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    Pilot #43 final CI + independent artifacts + PG17 root-cause boundary (2026-10-15)

    Run https://github.com/pgextwin/plpgsql_check/actions/runs/38746304977 SUCCESS; App15368 required Windows summary and PG14–18 jobs all SUCCESS, functional smoke, package, SPDX2.3 validate, Grype and upload steps SUCCESS. PR pgextwin/plpgsql_check#43 head 8d8b1452e7dcb61140040f96498aefae41e82536 remains OPEN (no merge). Downloaded all five Actions artifacts and independently opened inner ZIPs: ZIP CRC integrity PASS 5/5; DLL/control/SQL/metadata present 5/5; computed ZIP SHA256; SPDX-2.3 SBOM parses and its DLL SHA1/SHA256 checksums match ZIP DLL 5/5; Grype JSON parses, matches=0 in all five; PACKAGE-INFO buildInfrastructure.commit equals 4adc71076283dc9cf64543b760c65b054f5a85f0 in all five. PG17 detailed log: EDB installer 391.94MB download began 13:52:30.702Z and ended 14:03:14.228Z (643.5s), hashes matched 14:03:23.463Z, installer ran 14:03:23.568–14:05:35.495Z (131.9s). PG14/15/16/18 EDB download ~4–5s; installer execution ~116/133/150/131s. PG17 808s Install PostgreSQL step dominated by external EDB download wait, not observed installer execution regression. Underlying network/CDN reason and reproducibility remain unproven. Performance acceptance remains HOLD pending safe repeated measurement/defined external download mitigation; no manual dispatch authorized. No B2 modifications, signing, Publisher, Release, tags or settings changes.

  13. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    PG14–18 EDB install phase timing: minimal safe design (2026-10-15)

    Protected build/main 4adc71076283dc9cf64543b760c65b054f5a85f0; plpgsql_check pilot PR #43 remains OPEN/unmerged. Pilot run #38746304977 passed all PG14–18 Windows functional/build/SBOM/Grype/artifact checks. PG17 install-step 808s breakdown from completed Chocolatey job log: EDB 391.94MB download 13:52:30.702→14:03:14.228 (~643.5s); checksum completion 14:03:23.463; EDB install 14:03:23.568→14:05:35.495 (~131.9s). Other PG14/15/16/18 downloads ~4–5s; installer ~116/133/150/131s. Design decision: retain current Chocolatey invocation, --params, --ia, bounded retry and exit-code fail-closed unchanged. Avoid extracting/reimplementing EDB download or installer. Instrument phase boundaries from Chocolatey timestamped output in a separate read-only parser/report step (Download start/completed, Hashes match, Installing postgresqlNN, postgresqlNN has been installed); mark absent/ambiguous phase boundaries UNKNOWN rather than invent durations. Record attempt number, PG major, package version, installer URL basename/size, Download/Verify/Installer/Total durations, runner image, and artifact linkage without exposing secrets. Use independently testable fixture parser with PG14–18 log samples, retry/missing-marker/malformed-timestamp tests; avoid CI hard failure solely for missing timing telemetry. Acceptance: functional/ZIP/SHA256/SBOM/Grype and required App15368 CI must PASS; installer execution assessed independently of external download time; report network time and flag extreme outliers (e.g. PG17 download 644s) for follow-up, do not claim network root cause without evidence; compare multiple runs before declaring stable performance improvement. No implementation PR in this checkpoint: robust log capture and trust-safe telemetry need specification before modifying the shared installer; no workflow dispatch. Existing B2 13 PRs and release/signing/settings untouched.

  14. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    Issue #46 EDB phase timing implementation PR (2026-10-15)

    Created limited central build PR #111, latest head e207a9ece7a8ce539e43088a5b5bc610388d30c2, base protected build/main 4adc71076283dc9cf64543b760c65b054f5a85f0. Changes only 3 files: new read-only parser scripts/report-edb-install-timing.py, PG14–18 offline fixtures tests/test-edb-install-timing.py, and validation workflow invoking fixtures. Parser extracts timestamped Chocolatey log phase boundaries (Download/Verification/Installer/observed Total) and emits UNKNOWN for missing/duplicate/out-of-order/retry ambiguity; no Chocolatey installer invocation, flags, retry, exit-code or fail-closed logic changed. Scope: offline post-run analysis; not yet automatic telemetry emitted by every Windows run. Required App15368 resolver is pending after latest commit; PR unmerged. plpgsql_check Pilot #43 remains OPEN/unmerged, B2 13 PRs untouched. No dispatch, signature, Publisher, Release, tag or settings changes. Next: exact-head resolver CI, fix failures if any, then review whether automation of reporting is warranted separately.

  15. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    Issue #46 telemetry PR #111 safely merged; postmerge pending (2026-10-15)

    PR #111 merged with expected-head squash only after App15368 resolver SUCCESS in https://github.com/pgextwin/build/actions/runs/38751679492, exact PR head e207a9ece7a8ce539e43088a5b5bc610388d30c2, mergeable_state clean and protected main unchanged. Diff exactly 3 files: .github/workflows/validate.yml (+3), new scripts/report-edb-install-timing.py (+74), new tests/test-edb-install-timing.py (+43). No changes to Chocolatey install script, retry, flags, fail-closed. Required CI confirmed both minimal EDB contract and new PG14–18 read-only timing fixture test SUCCESS. Squash merge commit 0466d7275473cd88a95de650c4ac002c7941a67f; protected build/main now matches. Postmerge push Run https://github.com/pgextwin/build/actions/runs/38752126226 is QUEUED/IN_PROGRESS; App15368 resolver not yet confirmed SUCCESS. Tool is offline log parser only, not automatic Windows timing telemetry. plpgsql_check Pilot PR #43 and B2 13 PRs untouched/unmerged. No dispatch, signing, Publisher, Release, Tag, billing, Secret or permissions changes. Next: verify postmerge resolver SUCCESS before any downstream action.

  16. 16 remaining items

  17. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    #46 / #75 — Minimal EDB PostgreSQL installer acceleration: fleet stage 1 PRs created, HOLD_PENDING_PR_CI (2026-10-16 JST)

    User authorization: safely advance the existing validated pilot change across the 13 remaining extension repos, stage rollout through protected exact-head PRs and independently verified PG14–18 CI/Artifacts before merge; STOP on failure and HOLD when CI incomplete. Workflow Dispatch, signing canary, Publisher, Release, Tag, Secret/billing/protection settings changes prohibited. Nothing from B2 reimplemented.

    Latest source-of-truth:

    • central build PR #110 MERGED: immutable reusable builder commit 4adc71076283dc9cf64543b760c65b054f5a85f0; adds Chocolatey --ia "--disable-components pgAdmin,stackbuilder" without changing required PostgreSQL components, pinned package/retry/fail-closed checks.
    • plpgsql_check pilot PR #43 MERGED; postmerge Windows Run #38754020475 SUCCESS with independent PG14–18 ZIP CRC, SPDX SHA1/SHA256, Grype, PACKAGE-INFO evidence: prior pilot acceptance. Pilot is 1/14 existing caller pins.
    • Prior real pilot single-run per-PG phase metrics (not a statistically stable fleet-wide speed estimate): PG14 download 4.029s / installer 115.595s, PG15 3.507s / 133.363s, PG16 3.941s / 149.746s, PG17 643.526s / 131.927s, PG18 4.981s / 131.327s. PG17 anomalous external EDB download latency remains WARN; network/CDN root cause not established; mandatory integrity/functionality gates never waived. Prior overall Install PostgreSQL step PG14/15/16/18 was ~65–74% shorter vs old two-run mean, PG17 step much longer due to download. Detailed timing evidence.
    • Read protected main .github/workflows/windows.yml of ALL 13 remaining repos: each still pins old builder 455f0ccc667697f600700a6e64bcfd6c57b6a977, not optimized. Searched current open PRs: none implementing this builder pin upgrade. Existing unrelated PRs (candidate/version/Grype/deps) left untouched.
    • Keep immutable candidate-verifier ecf84241c1c4cd99fab3f964cb309f9c6e2a01f8 and operator-requested update-watch unchanged.

    Stage 1 limited to three non-overlapping extension repos (not 13 concurrent expensive Windows matrices): after independent protected-main/current blob/duplicate-PR checks, created one-line-only .github/workflows/windows.yml pin-change PRs, with no other files, settings or launch action:

    1. pg_bigm #31: expected PR head 71f2e4e21107bfd65889a918ad875a605241086a, original protected main/base 38d151d313c936115e286ab194f153db664bc5cd. PR Windows Run #38821619767.
    2. pg_cron #22: head 9513ce26c8ea759f5b2944bba5ad7b76f2bda0fd, original protected main/base c888c25737f47a18d0fac7369e03c28ad71b7a76. PR Windows Run #38821624982.
    3. pgaudit #19: head 8d75913af49b1e213e24c3df3bd914c9c160977f, original protected main/base cae087dcc6ffade4b18a63d272b3edcd23810dab. PR Windows Run #38821631205.

    Patch audited: 3/3 PRs exactly one file/one addition/one deletion: build-extension.yml@455f0ccc... → @4adc7107.... Check contexts are App ID 15368, protected main requires Windows required summary. Each pull_request Run automatically started (attempt 1, expected exact head); latest all IN_PROGRESS, Matrix Resolver SUCCESS, five PG14–18 Windows jobs IN_PROGRESS, no new Artifact yet. PR mergeable_state currently blocked due to incomplete required checks (not a conflict).

    Decision: HOLD_PENDING_PR_CI_AND_ARTIFACT_AUDIT, not a CI failure. No PR merged, no 10-repo follow-on rollout yet. When each stage-1 Run completes SUCCESS with App15368 required summary and PG14–18, download 15 new Run-bound artifacts and independently audit outer+inner ZIP CRC, SHA256, SPDX2.3 actual-file SHA1/SHA256, Grype JSON/database, PACKAGE-INFO exact source/Run and optimized builder SHA, functional smoke. Compare download vs installer time separately; unknown timing should be reported, never fabricated. Only then consider expected-head squash (rechecking head/base/protected main) and postmerge independent re-acceptance before rolling out next stage. Any failure => STOP and do not lower required checks.

    Remaining 10 unmodified caller repos: pg_hint_plan, set_user, pg_repack, pg_ivm, pg_qualstats, hypopg, wal2json, pg_partman, pg_stat_monitor, orafce. Pilot plpgsql_check remains accepted without rework.

    No manual Dispatch, signing, Publisher, Release, Tag, permissions, billing, Secrets or branch-protection changes.

  18. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    PG14–18 EDB minimal-install optimization fleet rollout — Stage 1 merged 3/3; remaining 10/10 PRs CREATED, HOLD pending CI (2026-10-16 JST)

    Authority: User requested introducing the already-piloted installer optimization into the last 10 remaining extension repositories, without weakening required CI, artifact trust gates, or forbidden publication/settings operations.

    Pin: pgextwin/build/.github/workflows/build-extension.yml@455f0ccc667697f600700a6e64bcfd6c57b6a977 -> @4adc71076283dc9cf64543b760c65b054f5a85f0 (central build PR #110; prior accepted plpgsql_check pilot PR #43). No other file or setting changed on the extension branches.

    First three Stage-1 PRs: independent PREMERGE acceptance PASS and exact-head squash MERGED

    • pg_bigm PR #31: App15368 required summary, resolver and PG14–18 Windows matrix all SUCCESS on premerge Run #38821619767; premerge outer/inner ZIP CRC 5/5, SPDX 2.3 5/5, 10 actual SBOM-listed SHA1/SHA256 checks match, Grype JSON database valid and zero matches, correct extension binaries and PACKAGE-INFO exact PR synthetic merge commit, Run ID/attempt1/ref and new immutable builder. Fresh exact head/base/protected main and mergeability CLEAN checked before expected-head squash. GitHub confirmed MERGED; current protected main 196aa94412c10ffd68e88cbfc1b5d1ac25449e88 and new pinned workflow matches. Normal push triggered postmerge Run #38822917135, initially QUEUED. Postmerge is NOT final accepted yet.
    • pg_cron PR #22: App15368 required summary, resolver and PG14–18 Windows matrix all SUCCESS on premerge Run #38821624982; premerge outer/inner ZIP CRC 5/5, SPDX 2.3 5/5, 10 actual SBOM-listed SHA1/SHA256 checks match, Grype JSON database valid and zero matches, correct extension binaries and PACKAGE-INFO exact PR synthetic merge commit, Run ID/attempt1/ref and new immutable builder. Fresh exact head/base/protected main and mergeability CLEAN checked before expected-head squash. GitHub confirmed MERGED; current protected main d4a3511083ab723264c361bef87e2476ef4a440e and new pinned workflow matches. Normal push triggered postmerge Run #38822927442, initially QUEUED. Postmerge is NOT final accepted yet.
    • pgaudit PR #19: App15368 required summary, resolver and PG14–18 Windows matrix all SUCCESS on premerge Run #38821631205; premerge outer/inner ZIP CRC 5/5, SPDX 2.3 5/5, 10 actual SBOM-listed SHA1/SHA256 checks match, Grype JSON database valid and zero matches, correct extension binaries and PACKAGE-INFO exact PR synthetic merge commit, Run ID/attempt1/ref and new immutable builder. Fresh exact head/base/protected main and mergeability CLEAN checked before expected-head squash. GitHub confirmed MERGED; current protected main 55f2aa65b77af925cacf6f928e5c384d6e98c5cc and new pinned workflow matches. Normal push triggered postmerge Run #38822937707, initially QUEUED. Postmerge is NOT final accepted yet.

    Stage-1 15 newly fetched premerge actual Artifacts independently audited, PASS 15/15, outer+inner CRC, SPDX2.3 and 30/30 file-hash checks; artifact/report JSON (local only) pgextwin-fast-installer-stage1-premerge-audit.json, report SHA256 f32add9f8d080fffd906bbceeee8449b24d294cd13039cf1064d1f3aea6252a5. Version-specific checks all PASS. Real EDB log phase timestamps (15 completed PG14–18 jobs) are single-marker per phase, no UNKNOWN: observed EDB download ~2.1–6.8 seconds; installer execution ~99.6–166.4 seconds; previous plpgsql_check PG17 download outlier ~643.5 seconds DID NOT repeat in these 15 jobs, but no claim of stable speed benefit or network root cause. This observation does NOT replace paired statistical performance comparison.

    User-requested remaining TEN extension PRs: CREATED 10/10, no merges yet

    Each of ten repositories: read existing protected main, required branch-protection Windows required summary App15368, current .github/workflows/windows.yml and its Git blob SHA, searched open PRs for duplicate installer pin upgrade, retained unrelated candidate/Grype PRs. Then created isolated ci/minimal-edb-pg14-18-20261016 branch from exact main SHA, changed ONLY one line in one file from old to pilot-accepted 40-hex builder SHA, opened PR with immutable/functional/artifact/CI gates and publication prohibitions. Retrieved complete PR per-file patch: ten of ten PRs verified exactly one file, +1/-1 line and no other change. All ten ordinary pull_request CI runs automatically started; at last observation all QUEUED or IN_PROGRESS with no completed overall conclusion. PR mergeable_state BLOCKED pending required checks, not a known conflict. All ten PRs intentionally remain OPEN/unmerged.

    Safety gate, current disposition and next priority

    • 14 extension fleet status: plpgsql_check prior final accepted; three first-wave extension main pins now updated but postmerge CI and Artifact acceptance pending; ten new extension PRs awaiting PR CI and independent Artifacts. Only 1/14 qualifies as full final acceptance at this checkpoint, 4/14 main branches point at the optimized builder.
    • HOLD_PENDING_POSTMERGE_CI for first three; HOLD_PENDING_PR_CI for remaining ten. Do not merge ten before exact-head App15368 required summary+all PG14–18 success and freshly downloaded five Artifacts/extension independently verified, and do not claim final acceptance of first three before postmerge SUCCESS and five new postmerge artifact checks.
    • For 10 PRs, perform each latest exact head/base/protected main and App15368 check, download 50 new PR-run Artifacts and independently check ZIP CRC, SPDX 2.3 actual DLL checksum, Grype valid DB/report, PACKAGE-INFO binding to new builder 4adc7107… and its PR-run packaging commit; examine per-PG install timing; if complete PASS then expected-head squash merges with subsequent postmerge CI and independent new Artifacts. Any CI failure/unsafe mismatched metadata => STOP; incomplete => HOLD.
    • No Workflow Dispatch, signing canary, Publisher, Release, Tag, Secrets, billing, permissions or branch-protection settings changed. B2 prior migration and prior plpgsql_check pilot untouched. No Release or signing acceptance is implied by optimized CI.

  19. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    Minimal EDB builder fleet expansion — incremental two-PR independent acceptance, HOLD (2026-10-16 JST)

    Further verification following stage 1 and ten new PRs:

    • pg_hint_plan #21: PR Run #38822321047 now completed/success. Five new PG14–18 Artifacts downloaded and independently audited; exact synthetic PR ref 5100d76214796ad7cd3718d4d55c887cb6dd2660 in PACKAGE-INFO matched GitHub PR merge_commit_sha; immutable optimized builder SHA 4adc71076283dc9cf64543b760c65b054f5a85f0.
    • set_user #19: PR Run #38822328953 now completed/success. Five new PG14–18 Artifacts downloaded and independently audited; exact synthetic PR ref 4b39ab0ef88739cc839bb92f0c0d64b7855abce7 matched; optimized builder pin matched.

    Independent file-level results: 10/10 artifacts PASS outer ZIP CRC, inner distribution ZIP CRC, valid SPDX-2.3, recomputed SHA1+SHA256 for each SPDX-listed packaged DLL (20/20 checks), valid Grype report with valid DB and matches=[], expected DLL/control/SQL and correct PACKAGE-INFO (extension, PG major, repository, exact PR merge-commit SHA, run ID/attempt1/pull_request ref, optimized builder SHA). Local read-only report pgextwin-fast-installer-stage2-premerge-audit.json, SHA256 3cdf69d934767120db53c2c92ae46efe4669822259e64772b1158f28ff4a812e — not GitHub-published.

    Safety: no merge of either PR or any other new ten was executed. All ten new PRs remain OPEN (2 premerge functional/artifact PASS; remaining eight CI QUEUED/IN_PROGRESS). The three earlier stage-1 PRs have been merged after independent PR acceptance, but their normal push postmerge Runs #38822917135, #38822927442, #38822937707 are all still QUEUED. Thus HOLD_PENDING_STAGE1_POSTMERGE_ACCEPTANCE plus HOLD_PENDING_OTHER_PR_CI; no fleet-wide completion claim. Resume by independently verifying three postmerge Runs+15 new artifacts, then rechecking any next exact heads and app15368 required CI before expected-head squash merges. Do not weaken or bypass gates for runner queuing.

    No manual Workflow Dispatch, signing, Publisher, Release, Tags, budget, Secret, permissions or protection changes.

  20. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    #46/#75 PG14–18 minimal EDB installer acceleration — fleet 14/14 main pin rollout COMPLETE; 10 postmerge Runs PENDING (2026-10-16 JST)

    Completed and independently accepted

    Fleet Stage2 10/10 PRs expected-head squash MERGED

    All ten PRs independently passed premerge GitHub Actions App15368 required Windows required summary + PG14–18 x64 + Resolver checks (correct head), main protection / exact head+base / CLEAN mergeability; one-line-only builder pin change. All 50 new premerge Artifacts were independently downloaded and validated for outer/inner ZIP CRC, SPDX2.3, actual DLL/EXE SHA1/SHA256, Grype JSON database/matches, PACKAGE-INFO exact synthetic PR commit, Run/PR ref, and builder SHA 4adc71076283dc9cf64543b760c65b054f5a85f0. No CI/binary finding. For orafce, an audit-manifest local ID-to-PG mapping mistake in PG15–17 was found by independent ZIP/package major checks, corrected by comparing GitHub authoritative Artifact names to actual ZIP contents and rerun; final artifact-identity/ZIP audit PASS 5/5, no repository artifact discrepancy. Combined separate audit covers Stage1 postmerge 15 and Stage2 premerge 50: 65/65 PASS, 140/140 SBOM file hash checks, local report pgextwin-fast-installer-comprehensive-audit.json and independent validator pgextwin-fast-installer-audit.py (local only; not GitHub published).

    Each expected-head squash MERGED=true, independently confirmed protected main at exact returned SHA; the latest workflow pin is the optimized builder. Ordinary push auto-triggered each new postmerge Run:

    Decision and non-negotiable next gate

    MAIN-PIN ROLLOUT COMPLETE 14/14; FINAL ACCEPTANCE currently 4/14. Remaining ten require independently confirming postmerge App15368 Windows required summary, Resolver and PG14–18 Windows matrix all SUCCESS at correct protected main; download 50 new postmerge Run-bound PG14–18 artifacts (do not reuse PR artifacts), verify outer/inner CRC, native binaries, SHA256/SPDX2.3/SBOM actual SHA1+SHA256, valid Grype report/DB, exact PACKAGE-INFO postmerge push Run/ref/main SHA and new pinned builder. Include read-only installer vs download timing where available and differentiate transient EDB download outliers. Until those are complete: HOLD_PENDING_POSTMERGE_CI_AND_ARTIFACT_AUDIT, not full fleet accepted. On failure STOP without bypass.

    Forbidden operations never performed: manual Workflow Dispatch, signing canary, Publisher, Release, Tag, Secrets, billing, global Actions permissions or protected-branch settings modification. Existing B2 and other unrelated PRs left untouched.

  21. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    PG installer fast-builder final main-pin confirmation; postmerge HOLD (2026-10-16 JST)

    Independently re-fetched protected main .github/workflows/windows.yml after all ten new squash merges. Every one of the ten now references exactly pgextwin/build/.github/workflows/build-extension.yml@4adc71076283dc9cf64543b760c65b054f5a85f0, none references former @455f0ccc667697f600700a6e64bcfd6c57b6a977, and all still pin unchanged verify-candidate.yml@ecf84241c1c4cd99fab3f964cb309f9c6e2a01f8. Together with previously accepted plpgsql_check and first-wave pg_bigm/pg_cron/pgaudit, 14/14 optimized builder pin deployed to main.

    Independent local consolidated record pgextwin-fast-installer-comprehensive-audit.json: Stage1 3-repository postmerge + Stage2 10-repository premerge Artifact evidence, total 65/65 PASS, actual native-file SPDX SHA1/SHA256 140 checks matched, full outer and inner ZIP CRC, SPDX2.3, Grype valid DB zero matches, exact PACKAGE-INFO source/build/run/ref and expected enhanced builder. This 65 does NOT include the ten new postmerge Runs; no attempt to substitute prior PR artifacts for postmerge acceptance. Earlier temporary orafce PG15–17 local audit ID mapping error corrected against authoritative GitHub artifact IDs and actual inner PG major, all ZIPs re-verified, no underlying artifact defect.

    Latest ten new postmerge run status: IN_PROGRESS — pg_hint_plan #38824715524, set_user #38824727887, pg_repack #38824739467, pg_ivm #38824751329. QUEUED — pg_qualstats #38824762778, hypopg #38824785987, wal2json #38824797838, pg_partman #38824808506, pg_stat_monitor #38824821188, orafce #38824832727. All runs are normal push attempt1 matching the exact protected squash main SHA. No final failure known; no new postmerge acceptance has been claimed.

    Decision: MAIN ROLLOUT PASS 14/14; FULL FINAL ACCEPTANCE HOLD 4/14 until last ten Runs complete SUCCESS and 50 new postmerge Artifacts independently pass same gates. No manual Dispatch, signing, Publisher, Release, Tag, billing/Secret/permission/protection changes. No further merges needed for this stage.

  22. ShutenOishi commented on Oct 15, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 15, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    #46 / #75 — Fast EDB builder postmerge final-acceptance STOP (2026-10-16 JST)

    Decision: STOP_FAIL_CLOSED — NOT 14/14 finally accepted. User requested read-only validation of the final ten normal-push postmerge Windows Runs with required GitHub Actions App15368 matrix and 50 new postmerge artifacts, and expressly required STOP on failure, HOLD while incomplete. All 14 protected main branches already pin optimized immutable shared builder 4adc71076283dc9cf64543b760c65b054f5a85f0; no new merge, Dispatch, re-run, repair, Release or setting change was authorized or performed during this acceptance turn. Existing accepted plpgsql_check, pg_bigm, pg_cron, pgaudit 4/14 untouched.

    Definite blocking failure — pg_hint_plan

    • postmerge Run #38824715524: completed / FAILURE, normal push, attempt1, expected merged protected main 312a89d3e6b55736c3d0eab61a175a1a92e084ee exact match. pg_hint_plan PR #21 was already merged; do not undo/bypass protection without new scoped decision.
    • App15368 Windows required summary FAILURE, because windows / PostgreSQL 18 / Windows x64 FAILURE. PG14, PG15, PG16, PG17 jobs SUCCESS; resolver SUCCESS. Four new PG14–17 Run-bound artifacts (IDs PG14 12014851550, PG15 12014413016, PG16 12014128991, PG17 12014323381) exist. No PG18 postmerge artifact (5/5 package completeness gate fails).
    • PG18 job ID 116729431101 inspection: Install PostgreSQL 18 SUCCESS; Build extension SUCCESS; Functional smoke test FAILURE. The smoke created a PG18 data directory, then pg_ctl could not start the server. PostgreSQL 18.6 log at 2026-10-15T23:41:21Z shows could not bind IPv6 address "::1": Permission denied, could not bind IPv4 address "127.0.0.1": Permission denied, FATAL: could not create any TCP/IP sockets; smoke script windows/ci/smoke-test.ps1:62 explicitly throws Failed to start PostgreSQL with pg_hint_plan preloaded. Packaging/SBOM/Grype/upload steps therefore SKIPPED for PG18. This is a functional acceptance failure, not evidence that PG installer itself failed. Exact underlying cause of loopback bind permission denied (runner OS permissions, port reservation, security software, or other) is NOT yet established. No bypass or weakened smoke is permitted.
    • Same PR head premerge Run #38822321047 PG18 job Install PostgreSQL 18, Build extension, and Functional smoke test all SUCCESS; nevertheless cannot substitute this earlier PR Artifact/run for failed postmerge head.

    Other nine postmerge normal-push run results (read-only)

    Completed SUCCESS with 5 new PG14–18 artifacts each from the correct Run (but not independently byte-verified in THIS turn due to STOP):

    IN_PROGRESS / HOLD: pg_partman #38824808506 (3/5 new artifacts visible), pg_stat_monitor #38824821188 (4/5 visible). These do not imply failure and cannot be marked accepted.

    All 10 Run metadata checked push, attempt=1 and exact postmerge main SHA. Existing historical stage-2 premerge 50-artifact CRC/SPDX/Grype acceptance remains valid for its own PR heads only; it does NOT replace postmerge audit. Because one required postmerge matrix and summary failed, stopped additional acceptance work before fetching/validating 50 new postmerge ZIPs or producing a new 50-artifact report. Thus: 14/14 immutable main pin deployed; 4/14 previously finally accepted; 7 new Runs succeeded but postmerge Artifact acceptance unverified; 2 runs pending; 1 failed. No all-14 acceptance claim or speed-reproducibility claim.

    Safe recovery proposal (NOT executed)

    1. Read-only investigate specific 127.0.0.1:55432 / ::1 bind denial and any differences with same-PR passing premerge PG18; inspect runner image, Windows excluded port ranges, local security/network state where visible; retain fail-closed functional test and required App15368 check.
    2. Separately decide a narrowly authorized ordinary GitHub Actions re-run or a test-only fix through a new protected PR after root-cause assessment; do not rerun without explicit scoped direction, never dispatch a signed/release workflow. A transient runner failure is possible but not proven.
    3. Once PG18 runtime smoke and whole required App15368 matrix SUCCESS at approved new evidence, independently fetch new postmerge PG14–18 artifacts for all ten and check outer/inner ZIP CRC, SPDX2.3, actual SHA1/SHA256, valid Grype DB/report, PACKAGE-INFO exact main/run/optimized builder. Measure EDB download vs installer separately with timestamped job logs; ambiguous telemetry UNKNOWN rather than invented figures. STOP on any integrity or required check failure.

    No operations violating user restrictions: no merge, Workflow Dispatch, signing, Publisher, Release, Tag, budget, Secret, global permissions, or branch-protection change in this turn.

  23. ShutenOishi commented on Oct 16, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 16, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    PG14–18 minimal EDB installer — pg_hint_plan postmerge PG18 socket-bind failure: GET-only root-cause investigation (2026-10-16 JST)

    Scope/safety: Read-only GitHub API, actual historical Windows job logs, PR patch, pinned reusable builder/metadata, official Microsoft/GitHub/PostgreSQL documentation. Only this Issue comment is a write. No Workflow re-run/Dispatch, PR edit/merge, release, tags, signed builds, Publisher, bill/Secret/permission/protection settings changed.

    Gate: STOP remains. pg_hint_plan postmerge Windows Run #38824715524, exact protected main 312a89d3e6b55736c3d0eab61a175a1a92e084ee, event push, attempt 1, completed/failure. GitHub Actions App15368 Windows required summary FAILURE, PG18 x64 job #116729431101 FAILURE; PG14–17 SUCCESS, resolver SUCCESS; only PG14–17 Artifacts uploaded. These failures remain blocking; do not substitute the premerge successful Run #38822321047.

    Direct failure pinpointed from actual job log

    • PG18.6 Chocolatey/EDB Install PostgreSQL 18 SUCCESS and Build extension SUCCESS; Install extension into test PostgreSQL SUCCESS.
    • Functional smoke test invokes windows/ci/smoke-test.ps1, using PGROOT=C:\Program Files\PostgreSQL\18 and PGPORT=55432, pg_ctl -D <RUNNER_TEMP>/pg_hint_plan-pg18-data -l <log> -o "-p 55432 -c shared_preload_libraries=pg_hint_plan" start. initdb completes, but server fails to listen.
    • PG 18.6 server log at 2026-10-15 23:41:21 UTC: could not bind IPv6 address "::1": Permission denied; could not bind IPv4 address "127.0.0.1": Permission denied; FATAL: could not create any TCP/IP sockets. smoke-test.ps1:62 throws Failed to start PostgreSQL with pg_hint_plan preloaded.; exit code 1. PG18 Package/SBOM/Grype/Artifact upload SKIPPED. There is no observed Address already in use or missing DLL/PG18 build error. Confirmed immediate failure = Windows TCP bind access denied on port 55432, not installer execution failure. Exact underlying OS-level denial cause unproven.

    Controlled premerge/postmerge comparison

    • Same PR: premerge PG18 job #116721704749 full smoke SUCCESS, with same PGPORT=55432, PG18.6 and optimized builder 4adc71076283dc9cf64543b760c65b054f5a85f0.
    • Premerge and failed postmerge identical smoke-test.ps1 blob 22b8b599874dbee746e05f9a89ccbf22999bdce1, and PR Step 10: disable Grype built-in kernel-header suppression #21 changes only .github/workflows/windows.yml immutable reusable builder pin (old->accepted fast pin); no smoke/auth/security option changed. Test contract remains v2 planner-hints/hint-table. pg_ctl uses default listen_addresses=localhost, which explains attempts on both ::1 and 127.0.0.1.
    • Both jobs: GitHub runner 2.337.0, Windows Server 2025 10.0.26100 Datacenter, windows-2025-vs2026 image 20261007.264.1, Hosted Compute Agent 20261002.596, same optimized builder SHA. Different VM worker identity, premerge westus, failed postmerge westus3. Same postmerge Run PG14 in westus3, different worker and same port 55432 SUCCESS; PG15/16/17 and multiple other repos PG18 also SUCCESS. Thus neither blanket PG18 incompatibility nor entire westus3 outage is supported.
    • Cross-repo comparative successful PG18 tests using same image/agent/port 55432: pg_bigm postmerge job 116723849073 (northcentralus), pg_cron 116723767123 (northcentralus), pgaudit 116723705695 (eastus), set_user 116729469926 (westus), wal2json 116730839169 (eastus2). No matching TCP bind permission error observed in these logs.

    Independent system documentation: hypotheses, NOT confirmed root cause

    • Shared builder metadata/postgresql.json pins testPort=55432 for PG14–18. This lies within Windows default outbound/dynamic port range 49152–65535 (Microsoft TCP dynamic ports).
    • Windows documents bind failures with WSAEACCES (10013) for denied/exclusive socket access and excluded/reserved ports; specific excluded-port behaviors in old Windows Server examples are not proof of same root cause on Server 2025. Microsoft Winsock errors, excluded port KB.
    • Rank hypotheses: (1) per-VM reserved/excluded port or TCP bind restriction at 55432 — strongest consistent explanation, still UNVERIFIED because failed ephemeral runner's netsh ... excludedportrange / port-owner snapshot unavailable; (2) transient exclusive port use or security/network filter — plausible but unverified; (3) PG18 defect or immutable builder regression — substantially weakened by successful prior same-PG premerge and other PG18 jobs. Same image version does NOT ensure same per-worker port state; same Azure region has another passing job.

    Recommended safe recovery (NOT executed)

    1. Prefer one narrowly authorized, one-time Re-run failed jobs for ordinary unsigned Run #38824715524, with no workflow_dispatch, signing, Publisher, Release, Tag, or settings changes. GitHub documents re-runs reuse original GITHUB_SHA/GITHUB_REF and original actor privileges (GitHub re-run docs). Only perform after fresh explicit user authorization and cost/artifact-gate check. Failed PG18 job and dependent required summary must return SUCCESS, with full functional planner-hint test retained. Important: GitHub re-run creates a later run_attempt and may leave PG14–17 artifacts from prior attempt; independently validate each actual artifact's attempted Run/provenance and the acceptance policy (do not silently label mixed attempts as same attempt or skip full five-artifact coverage). If the policy demands uniform attempt, use separately authorized full normal-CI re-run / new PR run instead; do NOT auto-accept a composite without its trust model.
    2. If repeated, or if per-attempt artifact provenance cannot be accepted: new protected diagnostic/test-only PR (not this turn). Add fail-closed, read-only pre-start diagnostics on Windows netsh int ipv4 show dynamicport tcp, netsh int ipv6 show dynamicport tcp, netsh int ipv4 show excludedportrange protocol=tcp, netsh int ipv6 show excludedportrange protocol=tcp, Get-NetTCPConnection -LocalPort 55432 -ErrorAction SilentlyContinue (optional). Capture whether port in excluded range (lack of LISTEN process alone is insufficient); record sanitized results. If changing smoke port, prefer a reviewed explicit test-port outside current dynamic/reserved ranges (e.g. evaluate 15432, only after checking host availability), and update versioned source/config so all pg_ctl, pg_isready, psql agree. Do not disable localhost/TCP verification, silence errors, globally change Windows port reservation/security or weaken required CI. Central metadata/postgresql.json currently provides 55432 for all five majors, so changing central metadata/re-pinning all 14 is a separate fleet-risk change, NOT a one-off implicit fix.
    3. After a clean required App15368 PG14–18 and summary PASS and valid PKG INFO, resume pending 50 new postmerge Artifact byte-level CRC, SPDX2.3 actual hashes, Grype, exact builder/source/run reference checks and download-vs-installer speed analysis; no full fleet acceptance before this.

    Disposition: STOP_CONFIRMED_PG_HINT_PLAN_PG18_LOOPBACK_BIND_DENIED; ROOT_CAUSE_UNDETERMINED_PORT_RESERVATION_HYPOTHESIS. 14/14 optimized builder pinned to protected main, 4/14 previously final-accepted. No security/required gates bypassed.

  24. ShutenOishi commented on Oct 16, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 16, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    pg_hint_plan ordinary Windows CI — owner-authorized one-time Re-run failed jobs STARTED; attempt 2 HOLD (2026-10-16 JST)

    Scope: owner explicitly permitted at most ONE failed-jobs-only re-run of unsigned normal Windows CI Run #38824715524, conditioned on fail-closed protected-main, run, App15368, storage/cost and permissions checks. No other workflows authorized, no release/sign/publisher/Dispatch/settings writes.

    Immediately before single API request: repository pgextwin/pg_hint_plan Public/unarchived; protected main exactly 312a89d3e6b55736c3d0eab61a175a1a92e084ee, branch protection requires Windows required summary App15368. Original Run #38824715524 was completed/failure, event push, workflow .github/workflows/windows.yml, head exact main, run_attempt=1, actor=triggering_actor=ShutenOishi. Failed exactly two jobs: windows / PostgreSQL 18 / Windows x64 #116729431101 (functional smoke bind-denied on localhost port 55432), and dependent Windows required summary #116731383291. PG14–17 and resolver succeeded. Exactly 4 nonexpired first-attempt artifacts, ID PG14 12014851550 / PG15 12014413016 / PG16 12014128991 / PG17 12014323381, total 198,390 bytes. No PG18 artifact.

    • Normal windows-latest on PUBLIC repo (GitHub documentation standard runner-minutes free); immutable optimized builder pin 4adc71076283dc9cf64543b760c65b054f5a85f0, normal builder GITHUB_TOKEN contents:read, no signing/attestation/publish step. Artifact storage retains billing/allowance applicability uncertainty (GitHub Free Org nominal 500MB shared), exact current headroom inaccessible to GitHub connector; no known newly discovered discrepancy; previous normal Windows runs and actual artifact uploads succeeded under existing settings. No budget or policy change.
    • Single write execution: GitHub API action rerun_failed_workflow_run_jobs called ONCE for exactly this Run, returned success=true. GitHub Run advanced to run_attempt=2, latest state in_progress, correct protected main unchanged. Only new PG18 Windows job #116738378442 is actually in progress. GitHub attempt-2 Jobs API also lists virtual/carry-over success entries for PG14–17 and resolver with original attempt-1 start/end timestamps; these are NOT freshly executed attempt-2 jobs, despite run_attempt=2 on those API job entries. The required summary is expected to re-evaluate afterward; do not infer its success.
    • Independent existing artifact verification: downloaded the four original PG14–17 postmerge artifact ZIPs fresh; outer+inner ZIP CRC, SPDX2.3 schema/data, recomputed native-file SHA1+SHA256 8/8, Grype valid report DB matches=[], PACKAGE-INFO exact source protected main, run ID #38824715524, attempt 1, push refs/heads/main and new pinned builder SHA all 4/4 PASS. Local audit pg_hint_plan-rerun-original-four-audit.json (not published). No PG18 result/Artifact yet; do not label 5/5 accepted.

    Decision right now: HOLD_RERUN_ATTEMPT_2_RUNNING, NOT SUCCESS. Await actual newly executed PG18 smoke result and App15368 required summary, inspect PGPORT=55432 server log on failure; if failure again, STOP no further re-runs. If SUCCESS, obtain new PG18 artifact and independently check ZIP CRC/SHA/SPDX/Grype/PACKAGE-INFO attempt=2 and original 4 attempt=1; distinguish mixed attempt provenance clearly. Historical first-attempt four remain valid individually but are NOT from attempt 2. An explicit run_attempt-sensitive publisher/signer contract is NEVER implicitly relaxed for this unsigned CI. If full Run status, required App15368 summary, five artifact bindings and mixed-attempt policy cannot all be justified, HOLD and request new scoped direction; no false 'all attempt2' claim.

    No second rerun, no Workflow Dispatch, tagging, signing, Publisher, Release, settings/Secrets/billing/permissions/protection change.

  25. ShutenOishi commented on Oct 16, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 16, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    pg_hint_plan STOP recovery — one owner-authorized failed-jobs-only rerun SUCCESS, independently audited 5/5 (2026-10-16 JST)

    Precise disposition: PASS_UNSIGNED_NORMAL_WINDOWS_CI_RECOVERY_WITH_MIXED_RUN_ATTEMPTS for pg_hint_plan only. Prior STOP_PG18_SMOKE_BIND_DENIED is resolved operationally by a clean single failed-jobs-only rerun. Root cause remains UNDETERMINED, no broad causal claim or direct registry/port change. Mixed attempts DO NOT authorize cryptographic signer/Publisher/GO A/Release and do NOT imply fleet 14/14 independently final-accepted. Fleet main builder pin remains 14/14 deployed and previously final accepted four extension repositories unaffected.

    Gate before execution

    • Public, nonarchived pgextwin/pg_hint_plan, protected exact main 312a89d3e6b55736c3d0eab61a175a1a92e084ee; Windows required summary branch-required App ID 15368; .github/workflows/windows.yml normal push, reusable unsigned builder immutable 4adc71076283dc9cf64543b760c65b054f5a85f0; workflow contents:read, signer/Publisher never invoked.
    • Run #38824715524 original completed/failure, event push, actor=triggering actor ShutenOishi, same protected head, run_attempt 1. Exactly two failed jobs: PG18 Windows x64 #116729431101 functional smoke TCP bind denied at 127.0.0.1/::1 port 55432; dependent App15368 required summary #116731383291. Four original PG14–17 artifacts exist; observed 198390 total bytes. Same class standard hosted runner in PUBLIC repo has free minutes per GitHub Docs, but exact shared artifact storage headroom and $0 hard stop applicability not fully observable through connector; prior normal CI uploads succeeded, zero changes made to budget/Secret/protection.
    • User explicitly authorized at most once Re-run failed jobs only for this Run. Called GitHub rerun_failed_workflow_run_jobs exactly once, API success=true. No duplicate invocation.

    Actual attempt 2

    • GitHub Run transitioned run_attempt=2 and ultimately completed/SUCCESS. App15368 Windows required summary job #116740767629 SUCCESS, PG18 Windows x64 actual new job #116738378442 SUCCESS, PG14–17 and resolver show SUCCESS in current check-runs.
    • Attempt2 jobs API also lists carried-over PG14–17 and resolver as successful with original attempt1 start/completion timestamps. Do not claim these were re-executed at attempt2.
    • Actual new PG18 job log: Windows Server 2025 image windows-2025-vs2026 20261007.264.1, region westus3, different worker from failed attempt1; PostgreSQL 18.6 installed successfully, pg_hint_plan compiled, PGPORT=55432 now server started at 2026-10-16T00:11:51Z. The functional smoke created extension, inserted 50,000 rows, applied query-plan SeqScan and IndexScan assertions without error (step SUCCESS), dropped table/extension and shut down server cleanly; no Permission denied/failed TCP bind line. Required Summary became SUCCESS at 00:13:50Z. No new code, port configuration, permissions or runner-image pin change.
    • The originally affected Windows host's reserved/excluded TCP ports were not logged; failure appears worker-local/transient but exact root cause remains unproven.

    5/5 actual GitHub Artifact byte-level independent audit

    • Downloaded all five GitHub Actions Artifact ZIPs from this exact Run; five are non-expired, correct Artifact name/PG major; single packaging protected main SHA 312a89d3e6b55736c3d0eab61a175a1a92e084ee; same immutable central build pin and workflowRun.id=38824715524, event=push, ref=refs/heads/main. PG14–17 original attempt=1; PG18 newly produced attempt=2, verified inside actual PACKAGE-INFO.json, not inferred from GitHub Run UI.
    • Each of 5/5 passes GitHub outer ZIP CRC, inner distribution ZIP CRC, safe paths, exact expected package identity and PG major, native DLL plus SQL/control, SPDX 2.3 valid fields & inventories, native file contents SHA1+SHA256 rehash 10/10 MATCH, valid Grype JSON/tool and DB status with matches=[] 5/5, PACKAGING SHA/Run/source and immutable fast builder 5/5. Zero detections are only for evaluated Grype DB/scan configuration.
    • Artifact IDs and computed independent hashes (outer Actions Artifact ZIP SHA256, then inner Windows distribution ZIP SHA256):
      • PG14 attempt1: Artifact 12014851550, outer 125f352fe42548f2b02f7a297277b1da424d8f867f003ecc9fa7139f142299aa, distribution 28afd2a24ee33b2d7ce5d3a22511af699699d3a1e3eaff7687ca5b08fc507a66
      • PG15 attempt1: Artifact 12014413016, outer ef21a6d956e79034a2823c3b2b1d4cdebe40c5b33910b9a895d694843216b935, distribution 31dba8acba99bdf47c3ffadaa895b6cad484474757d7118c6da7ef49b308d336
      • PG16 attempt1: Artifact 12014128991, outer 2d6a08294c10af32f9e96aa540e901987786e51b6f7e619bbd5a406f401fed7d, distribution a456d605a4b14849d1f7d0a1ca467d6d456d70faa614e1e820020f6c762ebf25
      • PG17 attempt1: Artifact 12014323381, outer 22a3c109efb6c169a0b3a84d7b681f24ff4781d0cf7296d141f439050642b896, distribution 5d7ca783f21d2f13fabff8676a058ff54e779395a22a702c786989e68de27840
      • PG18 attempt2: Artifact 12016181206, outer f53af826fa3cc014ed83e7106dfb5b0543479c2a00470b73a2886824f2fd4978, distribution 922da22d2982454dc61c8d28a68d5069de8740aece75ae67c15c7c25b7ae6163
    • Local report pg_hint_plan-mixed-attempt-audit.json, independent mixed-attempt validator pg_hint_plan-mixed-attempt-independent-audit.py, manifest pg_hint_plan-mixed-attempt-manifest.json: report decision=PASS, passed=5/5, failure_count=0, hash_checks=10, SHA256 of JSON report 3c78acbc8a1c9f80571464fd4056d066cac966bb4b052ed11bc15205d90055f1. No audit data was uploaded as GitHub artifacts; local files available to owner from conversation.

    Explicit risk/acceptance boundary

    • PASS for unsigned normal Windows CI validation and individual artifact integrity: GitHub overall Run & required checks SUCCESS, code/source immutable, exact Run+attempt per Artifact verified, no hidden replacement or bypass. Mixed attempts transparently recorded; combined artifact set is NOT a uniform single-attempt output.
    • NOT accepted for a signer/Publisher or cryptographic attestation gate requiring single run_attempt (especially attempt1); no formal Release eligibility or multi-extension 14/14 final audit inference from this scoped retry. Such a gate would need a separately reviewed uniform-run strategy and consent if later pursued.
    • No additional rerun/Dispatch, signing, Publisher, Release, Tag, PR edit/merge, billing/Secrets/permissions/protection setting changes.
    • Next separate workstream: remaining nine optimized-builder postmerge Windows Runs (previously seven success plus two in progress) must each have complete App15368 CI and own five fresh postmerge artifacts audited before claiming 14/14 fleet final acceptance. This one-run recovery does not inspect or accept them.

    Result: PASS_PG_HINT_PLAN_UNSIGNED_CI_RECOVERY_5_OF_5_MIXED_ATTEMPTS, ROOT_CAUSE_STILL_UNPROVEN, FORMAL_PUBLICATION_DENY.

  26. ShutenOishi commented on Oct 16, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 16, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    PG14–18 fast EDB installer — remaining NINE postmerge normal Windows CI FINAL PASS (2026-10-16 JST)

    Decision: PASS_9_OF_9_POSTMERGE_UNSIGNED_WINDOWS_CI. Nine protected-main postmerge Runs now ALL completed/success, event push, attempt 1, head SHA exactly current protected main. Each protected main requires Windows required summary App ID 15368; its checks + Windows PG14/15/16/17/18 and Matrix Resolver all SUCCESS from App15368, exact head. Nonrequired candidate verifier SKIPPED as expected.

    Independent audit of 45 freshly downloaded postmerge GitHub Actions Artifact ZIPs (NOT old PR Artifacts): exact repo/Run/PG14–18/Artifact ID, all nonexpired; outer Actions ZIP CRC 45/45, nested distribution ZIP CRC 45/45, duplicate/traversal path protection, appropriate DLL/EXE and extension files, SPDX-2.3 metadata/files 45/45, recomputed actual packaged file SPDX SHA1+SHA256 100/100 exact, valid Grype tool JSON & database 45/45 and matches=[], PACKAGE-INFO extension, PostgreSQL major, correct packaging repo+commit, workflowRun.id, attempt=1, push/main ref, immutable optimized pgextwin/build@4adc71076283dc9cf64543b760c65b054f5a85f0 45/45. Wal2json is a logical decoding plugin and its DLL, not control/SQL, is required. All 9 PASS, errors=0.

    Full machine-readable evidence (local to conversation, not GitHub-published):

    • pgextwin-fast-installer-nine-final-audit.json: SHA256 6ae88754ec9d6cd2885b0229a91295118ee138f863e799e9e3811a95d1344471. Per Artifact outer ZIP SHA256 and inner distribution ZIP SHA256, result of each integrity check, native SHA1/SHA256 count, exact GitHub Artifact ID/Run/packaging SHA.
    • pgextwin-fast-installer-nine-final-manifest.json: SHA256 59e385eba2f48503d717f10a9df0a3739f67e8ac1ac8bdbdd2a045f7c2010118.
    • pgextwin-fast-installer-audit.py: local independent validator; no GitHub modifications.

    Current GitHub retrievable Run/Artifact evidence (artifact IDs listed in PG14,PG15,PG16,PG17,PG18 order):

    • set_user Run #38824727887, protected main f840b029b1420e3509905e2a4a992e741b0adcf8, PG14–18 Artifact IDs: 12014522486 / 12014696640 / 12014143843 / 12014876572 / 12015071958
    • pg_repack Run #38824739467, protected main 54475cc7d8147846f7faf27ca8300b59b257c28f, PG14–18 Artifact IDs: 12015081782 / 12014686677 / 12014661615 / 12014967593 / 12015077072
    • pg_ivm Run #38824751329, protected main 3a9339d9bb976409c297538e442627972d6f1680, PG14–18 Artifact IDs: 12014562886 / 12014168737 / 12015086596 / 12014906430 / 12015251062
    • pg_qualstats Run #38824762778, protected main 5e50782c960008c08b354041cc3715197f7134ac, PG14–18 Artifact IDs: 12014793488 / 12014189738 / 12014318963 / 12014493653 / 12015231963
    • hypopg Run #38824785987, protected main 7d07cd6f95abe2ea4fe6aaba57df3fd8bb1c2c13, PG14–18 Artifact IDs: 12015052047 / 12014937991 / 12015381938 / 12014848475 / 12015236640
    • wal2json Run #38824797838, protected main 68b35bf34145163b0b8c5675207eb0b09398bffa, PG14–18 Artifact IDs: 12014523502 / 12014827304 / 12015216754 / 12014529002 / 12014972563
    • pg_partman Run #38824808506, protected main fc8f9a28a173e98b0a8ce81754f5f3865f30c7d6, PG14–18 Artifact IDs: 12014249846 / 12014904396 / 12015287105 / 12015074109 / 12014547582
    • pg_stat_monitor Run #38824821188, protected main f94f05fee80cd809f18657eb2c7c7c67900c2cc3, PG14–18 Artifact IDs: 12014319460 / 12015605050 / 12015620003 / 12015152939 / 12015068311
    • orafce Run #38824832727, protected main f400677ef5c7e33188f39be9c8ff39684d6ccdda, PG14–18 Artifact IDs: 12014259433 / 12014757498 / 12014223179 / 12014512906 / 12014987041

    EDB timing: independent actual 45 Windows job logs

    For each of 45 jobs, exactly one download start/end marker and one installer start/end marker identified and subtracted as two separate phases. All 45 have populated metrics (no UNKNOWN). Each PostgreSQL major n=9. Mean download seconds / installer execution seconds:

    • PG14: 4.394 / 129.633.
    • PG15: 4.929 / 136.455.
    • PG16: 4.920 / 128.050.
    • PG17: 4.959 / 179.998.
    • PG18: 5.788 / 132.900.
      Timestamp definition: Downloading postgresqlN 64 bit → Download of postgresql-... completed.; Installing postgresqlN... → postgresqlN has been installed. All 45 job IDs/durations and descriptive limits in pgextwin-fast-installer-nine-speed-metrics.json (local SHA256 44adc0eb77d890227cc36d67d864532eb99a01bd4ca42cb2c30bad84c77087e3). Pilot PG17 EDB download 643s outlier did not recur. PG17 installer is materially slower in this limited sample; exact mechanism not established. These are not before/after matched-randomized results; do not fabricate causal speedup percentages.

    Fleet final state and prohibited actions

    • 9/9 new extension final PASS. Previously accepted four plpgsql_check/pg_bigm/pg_cron/pgaudit intentionally not reworked; pg_hint_plan already accepted separately as unsigned CI recovery (Run #38824715524 with PG14–17 attempt 1 + PG18 attempt 2, mixed-attempt bound and independently verified) evidence.
    • 14/14 protected main pin deployed and 14/14 unsigned Windows CI acceptance accounted for, subject to explicitly identified pg_hint_plan mixed-run-attempt caveat. Not a homogeneous single-attempt set for signer/Publisher; no signing, attestation/release authorization or GO A/G4/G5 gates advanced.
    • This final audit was GET-only with local downloaded files and Issue evidence comments. No Workflow rerun/Dispatch, PR edit/merge, signing, Publisher, Release, Tag, billing, Secret, permission, or branch protection changes.

  27. ShutenOishi commented on Oct 19, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 19, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    2026-10-19 JST — Chocolatey selected-components fleet migration: 13/13 protected PR merges (postmerge validation in progress)

    Owner's specified method: Chocolatey, not portable ZIP. Reuses build PR #110 exclusions; build PR #139 is MERGED at protected main 474ae1745c5c570956b39de3d882febad560ea7f. Real independent benchmark Run 39192479767: PG14–18 5/5 MSVC build, SQL smoke and PG14/15/16/18 fresh Chocolatey install succeeded. PG17 reused preinstalled Windows runner version — not a measured fresh-install speedup.

    Production normal unsigned scripts/install-postgresql.ps1 uses unchanged pinned Chocolatey package/version and bounded retries, with EDB --enable-components server,commandlinetools --disable-components pgAdmin,stackbuilder --create_shortcuts 0 --install_runtimes 0. Other release workflow pins and existing signed formal Releases untouched. Pilot fresh installation timing: PG14 142.58s, PG15 161.02s, PG16 147.04s, PG18 159.16s; baseline regular installs previously 127–163s, so no statistically established wall-clock improvement yet. This is a verified minimal-component selection / installation correctness rollout, not a quantified speedup guarantee.

    All 13 extension PRs changed exactly one line in regular .github/workflows/windows.yml: normal reusable workflow full-SHA 4adc71076283dc9cf64543b760c65b054f5a85f0 → 474ae1745c5c570956b39de3d882febad560ea7f; independent latest main-file GET confirmed 13/13 pins. All original exact-head PR Windows PG14–18 5/5 plus App15368 Windows required summary SUCCESS before merge. All 13 expected-head protected squash merges succeeded:

    FINAL independent postmerge audit (2026-10-19 JST): ACCEPTED for normal unsigned CI fleet migration.

    • All 13/13 extension protected-main exact-merge push runs are COMPLETED / SUCCESS, attempt=1. Run IDs and individual links are in the inventory above, and each head_sha matches the independent protected main commit GET.
    • 65/65 real Windows PostgreSQL 14–18 matrix jobs SUCCESS (13 extensions × 5 maintained majors); 13/13 required App15368 Windows required summary SUCCESS; 13/13 windows / Resolve extension and PostgreSQL matrix SUCCESS. Candidate-only SKIPPED jobs are never counted as success.
    • Independently fetched and inspected all 13 run job summaries after their parent Runs completed. Zero failed matrix jobs, no CI waivers or manual re-runs. All 13 current main .github/workflows/windows.yml pins match 474ae1745c5c570956b39de3d882febad560ea7f, all 13 main refs report protected=true.
    • The earlier plpgsql_check pilot was also merged (PR #47) and its postmerge Windows CI SUCCESS. Thus 1 pilot + 13 additional = 14 extension caller repositories migrated.
    • This acceptance covers reliable selected-component installation + build/test correctness, not a proven absolute reduction in installer wall-clock seconds: fresh pilot 142.58–161.02s, historical normal reference 127–163s. PG17 runner preinstall reuse is not a fresh-install benchmark. Keep this distinction; consider a separate matched-run performance analysis if speed metrics are required.
    • No formal release/tag/new publisher, no promote.yml or Signer-pinned release path change, no budget/secret/protection relaxation. Issue Eliminate Work-only Dispatch: Chat-authorized Release trigger and automatic Catalog proposal #75 GO B/G4/G5 remain distinct and unaltered. Original Issue CI blocker: Chocolatey HTTP 504 during Windows PostgreSQL installation #46 transient Chocolatey feed 504 concern is not declared eliminated by a single successful fleet run.

  28. ShutenOishi commented on Oct 19, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 19, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    2026-10-19 JST — FINAL: Chocolatey selected-components fleet rollout postmerge accepted 13/13

    Follow-up to interim report. Independent authenticated GET confirms all thirteen extension main push workflow runs are completed/success, event=push, attempt=1, with exact expected merge SHA. Across all thirteen runs, each PostgreSQL 14, 15, 16, 17, 18 actual Windows x64 job succeeded (65/65 total), and each mandatory App15368 Windows required summary succeeded (13/13). No failed runs waived; candidate verifier SKIPPED jobs are not counted.

    Current protected main fetched directly for all thirteen repositories: 13/13 protected=true and 13/13 normal .github/workflows/windows.yml pins equal protected build/main 474ae1745c5c570956b39de3d882febad560ea7f; independent pilot plpgsql_check postmerge 39193491773 also SUCCESS and uses same SHA. Thus the one pilot + thirteen extension normal unsigned CI rollout is COMPLETE and CI-ACCEPTED, 14/14.

    Change scope: SHA-pinned Chocolatey PostgreSQL installers now select server,commandlinetools, preserve proven pgAdmin,stackbuilder exclusion, skip shortcut/VC runtime reinstallation, maintain required executables/headers/import libs, fail-closed retry and actual SQL smoke. Pilot original-byte stopwatch: PG14 142.58s, PG15 161.02s, PG16 147.04s, PG18 159.16s; PG17 reused runner preinstall and was NOT a new-install speed benchmark. No measurable universal elapsed-time speedup has been independently established; acceptance is functional deployment + CI success, not a numerical performance claim.

    No release workflow/Signer pins, tags, formal Releases, Catalog, Website, Secrets, branch-protection or billing were modified in this rollout. Issue #75 GO B/G4/G5 are separate and remain unchanged.

  29. ShutenOishi commented on Oct 19, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 19, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    FINAL ACCEPTANCE — 2026-10-19 JST: Chocolatey component-selection fleet rollout COMPLETE (14/14)

    Disposition: PASS_FLEET_14_OF_14_EXACT_MAIN_POSTMERGE_WINDOWS_CI. Scope is normal unsigned CI installer, not signed release promotion or a quantified speedup assertion.

    Closure interpretation: original external Chocolatey v2-feed HTTP 504 availability is not magically fixed; bounded retry/fail-closed behavior and the actual fleet CI blocker are operationally resolved. Future 504 incidents get new reproducible evidence, not indefinite hold on an already tested rollout. Historical failures remain failures.

    No Release/tag, Signer/Publisher workflow, Secrets, org budget, branch-protection, Catalog record or Website publication changed. #75 GO B/G4/G5 and #112 signed distribution gates remain separate and unaffected. Close this implementation/CI incident as completed, not as proof of speedup.

  30. ShutenOishi
    closed this as completedon Oct 19, 2026

  31. ShutenOishi commented on Oct 19, 2026

    @ShutenOishi
    ShutenOishi
    on Oct 19, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    Final PG14–18 fleet installer step timing — exact protected-main postmerge audit (2026-10-19 JST)

    Final rollout: COMPLETE (14/14); Windows PG14–18 postmerge 70/70 SUCCESS; mandatory App15368 required summary 14/14 SUCCESS. Each current protected main was independently re-fetched and matches the exact postmerge run SHA; 14/14 branch protected=true. Verified Chocolatey server+CLI selection, no release/Signer/Publisher changes. Build PR #139 and caller protected PRs were previously MERGED.

    Actual installer workflow-step elapsed seconds, not entire matrix job duration. GitHub Actions original REST /actions/runs/{id}/jobs?per_page=100 per step started_at→completed_at, in UTC ISO seconds. Step is Install PostgreSQL {major}; includes installer invocation, Chocolatey feed/network/retry as applicable and postinstall mandatory component checks; values are not pure EDB executable processing time. All shown current step statuses SUCCESS on push/main attempt 1.

    Extension (exact-main postmerge Run) PG14 PG15 PG16 PG17 PG18
    plpgsql_check 155 177 121 229 223
    pg_bigm 164 162 142 225 184
    pg_cron 158 144 142 191 154
    pg_hint_plan 153 119 157 185 142
    pgaudit 144 130 157 203 151
    set_user 130 165 168 221 150
    pg_repack 134 152 135 204 153
    pg_ivm 148 151 139 185 159
    pg_qualstats 143 147 118 219 123
    hypopg 150 142 155 200 141
    wal2json 144 136 155 211 155
    pg_partman 128 164 171 167 166
    pg_stat_monitor 125 166 164 165 163
    orafce 168 143 160 201 162

    Old→new per-major 14-repo single-run mean step seconds:

    • PG14: 158.3 → 146.0 s (observed -7.8% step time change)
    • PG15: 170.2 → 149.9 s (observed -12.0% step time change)
    • PG16: 156.6 → 148.9 s (observed -4.9% step time change)
    • PG17: 204.0 → 200.4 s (observed -1.8% step time change)
    • PG18: 165.6 → 159.0 s (observed -4.0% step time change)

    All 70 old samples total 11,966 seconds (mean 170.9 s) vs all 70 current samples 11,258 seconds (mean 160.8 s): observed aggregate -5.9%. Of 70 pairs, 43 had lower step elapsed and 27 higher. Per-repository prior protected-main workflow run IDs: plpgsql_check 39183736517, pg_bigm 38822917135, pg_cron 38822927442, pg_hint_plan 38824715524, pgaudit 38822937707, set_user 38824727887, pg_repack 38824739467, pg_ivm 38824751329, pg_qualstats 38824762778, hypopg 38824785987, wal2json 38824797838, pg_partman 38824808506, pg_stat_monitor 38824821188, orafce 38824832727.

    Interpretation / evidence limitations: One old and one new runner execution per PG/repository, not controlled paired runs on identical Windows runner installations. Runner preload/cache/CDN contention/Chocolatey retry may differ. In particular the old pg_hint_plan Run 38824715524 was a mixed attempt-2 workflow and is not homogeneous with other baselines. Thus the -5.9% is the observed step-time difference across 70 pairs; it is not proof that Chocolatey selection alone causes that reduction, nor a guaranteed future speedup. The verified operational completion is 14/14 deployment + 70/70 functioning postmerge CI.

    No new workflow Dispatch, PR edits, release/promote, bypass, permissions/Secrets/budget/branch-protection changes were made for this timing audit. Issue #75 GO B/G4/G5 and independent signed-release gates remain unchanged.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions