Repository navigation
Conversation
Restore the guarded file-route drain and exact stop ownership boundaries from lidge-jun#8803 after its Windows timeout revert. Replace repeated CIM owner probes with same-handle native creation/SID evidence, preserve Windows argv semantics, retain apply exclusion during recycle, and retry only closed, unpublished rollback listeners within a shared remaining budget. Native Windows acceptance remains pending; no CI policy or budget changes.
|
Important Draft PR not reviewedDraft PRs are not automatically reviewed by default.
To automatically review draft PRs, update your CodeRabbit configuration: reviews:
auto_review:
drafts: true
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. |
|
✅ Deterministic PR hygiene checks passed. |
Apply the identical generated timing table from lidge-jun#8940 (d190515). Completed primary-file samples update scheduling weights only; test coverage, five-shard execution, time limits, and workflow policy remain unchanged.
Preserve upstream same-handle argv reads, Hub incarnation checks, relay refresh cleanup, volatile lifecycle locks and fixture phase deadlines. Keep strict native owner evidence, Windows argv parsing, rollback retry ownership and Link recycle exclusion from this follow-up. Use the reviewed upstream-preserving timing table from lidge-jun#8940 (e584532). Native Windows acceptance remains outstanding; keep this PR in draft.
Preserve the 26 historical evidence snapshots and append the current CI fingerprint set with the independently reviewed timing table blob. Keep every workflow, job requirement, shard count, and timeout unchanged. Validation: 69 release evidence/history/dispatch/resume/docs tests pass; release-evidence policy-check passes.
Preserve the existing PR work while merging current dev. Keep five shards, individual test and batch deadlines, required checks, triggers and release identity rules. Append the reviewed workflow fingerprint without changing prior policy records. Refs lidge-jun#8940
…ut shard Move the existing windows-stop-owner-identity mapping next to the Link stop recovery tests. This restores the central layout to the 1,999-line dev blob while preserving the Windows test domain and its independent fixture. On Windows x64 with Bun 1.4.4 the repository file-size ratchet, both layout suites, 2,683-file typecheck and structure check passed. Native Link recovery still reaches its documented 15-second watchdog. Keep this PR in draft; the full suite was omitted for this scoped manifest update.
|
The original eight-file Windows run at On a physical Windows host at current head Three representative cases in |
|
I ran the first failing Windows recovery case on both the current base (
On both revisions recovery returned 409 while fenced, then 202 after a completed disconnect. The fixture observed one replacement listener, preserved the user setting, and verified that the connection token and old runtime record were gone. The replacement was the fixture’s loopback listener. These runs did not start an installed service. I also ran the pending-disconnect case with the same diagnostic allowances. Both revisions still returned 409, retained the client role and runtime record, and started zero replacement listeners. In a separate instrumented PR run, the disconnect phase spent 3.20 s in a synchronous Windows owner/DACL check. Sixty-three permission-hardening command steps from atomic config writes took another 2.23 s in total. Those are measured costs on this guest; one sample per revision does not establish a relative slowdown. The next place to look is repeated config writes during disconnect, preserving ownership and ACL checks. The first stock failure therefore reproduces on the base. The original deadlines and the other failures in the eight-file Windows run remain open. All timing changes were confined to disposable test copies. |
|
Windows follow-up at The scheduler now defers a timer-only sync for five seconds when the published PID, port, attestation identity, generation and stopping marker match. A Home-requested sync still takes the ordinary guarded path and receives its original result or error. The existing checks remain authoritative whenever a sync runs. In a traced run of this head the periodic tick deferred, the timer-originated owner check disappeared, and the real checks at disconnect and recycle still ran. On Windows with Bun 1.4.4, all five scheduler tests and the runtime publication/teardown test passed, including foreign and malformed listener records. The original 15 s recovery case still failed at 15.04 s. With diagnostic deadlines the completed-receipt case passed in 21.35 s and the pending-receipt case correctly refused a replacement. Keep this PR in draft. The test harness now prints timed phase transitions on failure and records when its own deadline requested termination. Separate controls confirmed that an unrelated exit 137 is reported as a child exit, and only the exact bind-collision diagnostic is retried.
For an independent reference, the hosted Windows shard at Same-guest AB/BA on the current head. I ran unmodified
In one PR run the child received restart acceptance at 14.86 s but did not report completion before the deadline. In the other it last reported disconnect. The terminated runs cannot measure relative completion speed. Using a private transpiler cache per revision, including an empty run and a reuse run with the CI harness's Bun flags, did not change the stock bound-restart outcome on either revision. A separate three-sample check on empty synthetic directories compared the same trusted PowerShell executable with the same successful strict ACL verifier. Median launch-only / verified directory times were 2.106 s / 2.218 s in the sandbox and 0.340 s / 0.422 s on its physical host. These timings cover individual operations. Physical Windows follow-up, Bun 1.4.4. The same positive bound-restart test, with the stock deadlines and a fresh test home for each run, stopped at 15.05 s on base On the PR, a diagnostic copy extending the outer test and parent timeouts completed in 17.23 s. The child’s 10 s watchdog after listener startup, the real ACL checks, and the recovery assertions were unchanged. The child started disconnect at 9.74 s, requested restart at 15.97 s, received acceptance at 16.34 s and began binding the replacement listener at 17.16 s. A separate timing-only copy measured 0.43 s in the preliminary orphaned-enrollment check, 1.77 s in file-route cleanup and 3.46 s in disconnect finalization. Seven recovery-receipt writes within finalization took 2.33 s combined. Their time is included in the 3.46 s. Removing the preliminary check alone would not recover the observed gap. The stock deadline remains an open Windows gate. Follow-up #8996 isolates an optional Windows ACL performance change. On the same source tree combining this head and #8996 ( |
The recovery fixture now sends a missing creation-time sample through the same ownership decision used by normal samples. A missing identity must keep recovery fenced. A matching or mismatching observed identity retains its old result. Windows x64 with Bun 1.4.4: five isolated decisions checked before and after, and typecheck passed. The parent-unverified integration fixture reached its unchanged 15-second parent deadline during file-route setup, before the observer ran. Its Windows integration result remains open.
|
Windows verification at With phase timestamps added to the test, the stock A separate diagnostic used the unmodified Child and production source. Only the parent observation deadline (30 s) and the enclosing test timeout (35 s) changed. The Child's 10-second watchdog after listener startup and the original assertions remained in force.
The longer parent observation exposed a second deadline in the latter two cases. No checked-in timeout or operational setting was changed for these measurements. |
|
Skipped in the current maintainer integration batch. The description's Outstanding Windows gate is still open (the bound-restart and pending-disconnect fixtures reach the 15 s watchdog on physical Windows), and PR CI runs only on Ubuntu, so this batch can't verify it. Note that #8931 and #8974 have since merged into |
Summary
Follow up on the Link stop restoration now merged into dev as #8937. The current diff retains the remaining Windows command-line, native ownership, rollback and recycle corrections. Saved connections and credentials remain intact; conflicting or unverified ownership refuses destructive actions.
Outstanding Windows gate
Draft. Two native Windows cases from #8930 passed in earlier testing. On a physical Windows host with Bun 1.4.4, the positive bound-restart case exceeded the stock 15-second watchdog on both
456946c5and based61cca12. The isolated stop fixtures at current head2d3a18a50still exceed that deadline. The wider Windows recovery gate remains open.Verification
Current head
2d3a18a50Previous head
456946c5Earlier verification by commit
Native Windows verification on
8e4d926feff87debfed6e2042154c7d0632cf369The merge incorporates
e6432dbc6e75551c1135b2c80451290cf9d4b19dfrom dev.git merge-treealso produced clean results against dev5f80219ff91e273512b2fbb09374205844122777and #8931 head9dbc668857eba41cc2b274211a3647c85679b97e. The runtime diff following #8937 still includes the same-handle command-line/owner evidence, rollback lock handling, and recycle exclusion through teardown.scripts/test.ts --no-env-file --no-install --isolate --parallel=1. The Windows stop-owner identity, recycle exclusion, rollback lock and two layout suites passed all 48 tests. The 30 native in-process creation-proof test passed in 2.55 ms.bound restartrecovery case was also isolated. The untouched fixture exited 137 at 15.06 s on this branch and 15.05 s on unmodified dev5f80219f, with the same host, Bun, dependencies and real ACL operations. In a separate instrumented run, startup finished at 9.35 s, preflight took 1.55 s and disconnect started at 12.47 s before the 15 s parent watchdog terminated the fixture. A diagnostic-only control using the repository's Windows ACL subprocess stubs completed in 3.71 s. The PR keeps real ACL checks and the original deadlines. The isolated comparison covers this case only.bun run typecheckpassed for 2,654 files.bun run structure:check, the release-evidence policy check andgit diff --checkpassed. PR CI for this head. The full local suite was omitted during shared Windows validation.Earlier dev integration and CI job budget — 2026-11-19
Head
d04a6d0ed5db477e7e11f56c42b4e325962e0adc, tree96b3ffdec72d04d036d5503d8195accbd8b943e5, with normal-merge parents8e8623bb78d041748aca876a26ea55b0fa182437and dev676baf650c2d2b020439341f65e1ccc57c5d3e90. GitHub's actual PR mergea69ebd44b1372d0e6aa757aaeb0f93950d713b98has this same tested tree. The PR-owned runtime and test files (13 files) remain byte-identical to the previous PR head; upstream features and the layout/structure inventories are retained.The common change from #8940, source
82ee1a5a483c3e4e7756d5cb000f72ab98e376a5, gives the five full Ubuntu shard jobs a bounded 18-minute outer job limit, previously 15 minutes. The prior runs were cancelled near the 15-minute limit while tests were still progressing or finishing. The per-test and per-batch deadlines, five-shard coverage, required jobs, workflow triggers, permissions, trusted-dev/manual-dispatch rules and release identity checks are unchanged. This is an outer CI-job budget change, not a passing result for the earlier cancelled runs.Current dev's timing table
a07f90e6098082e1452ea5cc075c801271321ecais preserved. The release policy retains all 27 prior records and appends one exact workflow snapshot. Reviewed blobs: workflowda0b271b0d8c362a3efe98485ebfc543f64b03a2, strict budget testb8fa9b20e23a0ed571c2ded0b655f22492ad43f6, and policy7df235ec7904dd19a4b2e588ecfba1da627e9972. Independent source and security review covered the immutable common patch3e18ab649f9766f74859c7eb9ae2f400f1eddbc8d8c4496f77a22fff5f675f7b.Validation on this integrated tree, using Bun 1.4.4 on Linux:
scripts/test.tsrunner with--no-env-file --no-install --isolate --parallel=1ranwindows-stop-owner-identity,client-recycle-exclusion, andclient-restart-rollback-lock, together withci-workflows, both test-layout files, the file-size ratchet and Spanish localization: 245 pass, 1 skip, 0 fail; 5,801 assertions across eight files. The one skipped check requires native Windows. This PR remains draft with the existing Windows acceptance requirement.node .github/scripts/release-evidence.cjs policy-checkpassed.bun run typecheck: 2,632 files, no errors.bun run structure:checkandgit diff --cached --checkpassed.Exact-head main CI 44472792150 is queued as of 2026-11-19 06:33:30 UTC. It has not established a passing result. No manual workflow dispatch or additional retry was started. The full local suite and documentation build were not repeated for this integration during shared-worktree activity; earlier runtime/docs results below remain evidence for their explicitly named trees. Native Windows acceptance and the existing review requirements remain separate.
Previous release-evidence follow-up — 2026-11-19
Head
8e8623bb78d041748aca876a26ea55b0fa182437, tree69ae71fcb44374e26f3ac981a71c811515f1c202, direct parente3c6154dde46ad1f07d0a0af93c4fc4626cc6ff6.The reviewed timing table from #8940 also needs an exact fingerprint in the release-evidence registry. This follow-up changes only
.github/scripts/release-evidence-policy.json: all 26 historical records are preserved, and one snapshot is appended with the existing CI requirements and timing-table bloba07f90e6098082e1452ea5cc075c801271321eca. The approved policy blob is51c1f898f2bdd69b7a012bd191ac51acb3c1779b; it received independent review. Workflow triggers, required jobs, shard counts, time limits, and test coverage retain their existing values.Validation on this exact tree:
release-dispatch-guard,release-evidence,release-history,release-resume, anddeploy-docs).node .github/scripts/release-evidence.cjs policy-checkpassed and confirmed that the registered fingerprints match the working tree.git diff --cached --checkpassed. Before this snapshot was added, the timing-table fingerprint assertion failed.Historical run 44463011731 for this parent head finished cancelled. Shard 4/5 completed all 63/63 batch summaries, with no failing assertion in its full log, before cancellation. The run was not a passing result.
Previous dev integration — 2026-11-19
Head
e3c6154dde46ad1f07d0a0af93c4fc4626cc6ff6, tree615e902f66b3ba183937a28371e97136812b1119. This is a normal merge of9914824e9a9b90280f532ee1a5eb7c5a61be78cband dev80e8952357d11ccf1421b4aa5620e781625dcdb0, published with an exact-head fast-forward lease. The merge preserves the source changes from both branches; it does not restore the older implementation over the maintainer's changes.Bun 1.4.4, the normal
scripts/test.tsrunner/preload,--no-env-file --no-install --isolate --parallel=1 --timeout 30000, and a private test temporary filesystem were used:windows-stop-owner-identity,client-link-stop,client-link-stop-recovery,client-recycle-exclusion,client-restart-rollback-lock,stop-deferred-teardown, andwindows-process-evidence: 288 pass, 5 skip, 0 fail; 2,661 assertions. The portable owner-enrichment test was then run with PowerShell 7.7 on PATH: 1 pass, 27 assertions. Four native-Windows tests remain skipped.client-hub-relay,client-runtime-restart,client-link-stop-sibling, androute-intent-runtime-stop: 163 pass, 2 macOS skips, 0 fail; 1,337 assertions.5920933f76e412ffa05d3b61d3aa28de21308b72360195e34d752c943700db0dfound no remaining implementation blocker. It checked the recycle admission/teardown boundary, listener closure before retry, owner classification, same-handle SID/liveness/creation/argv checks and preservation of Hub refresh cleanup.e5845320b776a27aa58562f80e8d968140bc9046; table SHA-2562d56880fe64338d36b424bb73178b025078ca1482da5dd4be6e73c6792759be5. Completed primary-file samples from dev run 44386662799 and Ubuntu PR runs 44448708478/44448420609 are used; the cancelled workflow and unfinished sample are explicitly distinguished. All 20 newer dev rows are preserved. Independent replay of the actual runner retained all 3,358 general files exactly once across five shards. The common candidate's 42 harness tests / 6,670 assertions are scoped data/runner evidence, not this PR's full-suite result.460036b66ee9d8eac99af25ced81010fa8c5676f, and Service/Catalog/PR gates are queued. No queued check is described as passing.The first local run refused the history coordinator because the temporary namespace had the wrong owner/mode. It was stopped; the successful runs above used a private root-owned
1777temporary namespace without changing host permissions or production ownership guards. Full local repository tests were not repeated during concurrent worktree activity. Native Windows, the two macOS-only handoff cases and full exact-head CI remain outstanding. The earlier results below belong to their earlier tree.Earlier validation before dev integration
Candidate based on
d4a0382ceab624c44a7753d50054379f57933ad4; reviewed tree:5609284b6df4842fa45f95bf9fced54147524e68.bun run typecheck: passed, 2,592 files.bun run structure:check: passed;bun run privacy:scan: passed;git diff --check: passed.ASTRO_TELEMETRY_DISABLED=1 bun run buildindocs-site: passed, 775 pages and 117,969 internal links.4ffd5824e8e0e7e2f1ea6f1c8e3aeed65df16840a997aba5657d78a52c17f80a; the immutable follow-up-only review patch isff1fb76bc8bca2189992ee23725e87aeb662d1d880f16b428a5c1c159e6c8608.Checklist
Review readiness
456946c5includes based61cca12(11 commits ahead, 0 behind at verification).