Repository navigation
Diagnose and bound scoped GitHub inbound reads before remote closure projection #256
Description
CONTINUATION-SCOPED-INBOUND553-TIMEOUT-20261011
User continuation via Willman encountered a reproducible transport stall on current published SDKf0a61b05006c948150955d2a449ad2202a25048e/0.1.126. Existing PLF113 providerfailure handling and PLF13804 CAS source work are delivered; this is a fresh bounded diagnostic request, not a competing implementation. ParentWillmanPLF17303/Issue193. Exact configuredcommand in subactor/onedev-agent: planfile sync github --direction from --ticket PLF-17307. It prints Connected and Found1ticket, then provides no closureprojection/result within45s realWillman taskba2f1e91392b44a19cb61fd108e2bead; one retry also exceeded120s and root terminated only its exact owned CLI PID after verifying argv/cwd/startTicks/stdout target. Actual remoteIssue553 isclosed at20:08, localticket remainsopen. NativeSDK/runtime canary itself ishealthy888PASS/160subtests and is not blocked by this bookkeeping transport. Remoteclosure preserved; outbox retained, no repeated outboundreopen or sourcechanges. Cause is not yet localized; do not infer credential/provider/code failure from silence alone.
Acceptance: owning native diagnostic/allocation/currentsourceadmission and independentvalidation. Identify exact blocked stage, per-ticket provider pagination/cache/coalescing and timeout behavior; bind deadlines to monotonic operation lifetime, not fileactivity; verify --ticket selects bounded linkedremoteintake. Timeouts must return explicit nonzero/structuredpartial effects, preserve localCAS/currentleases/sourceidentity, previouscomplete artifacts and remote terminalstate; no success claims or globalimport/forcedstatus bypass. Add intercepted slow/hanging provider and normal exactticket closure controls through publiccommand before independent current-head/base CI. Source fix/runtimeactivation require their ownprotectedscope; this queued request grantsnone.
Evidence: Willman .subactor/receipts/autonomy-20261011/continuation-pr227-actual-publication-queue.json, continuation-pr227-actual-transports.json, continuation-pr227-inbound553-retry.log, continuation-pr227-inbound553-owned-process.json; retained .subactor/recovery/autonomy-20261011/pending-onedev553-inbound.json.
Strategy Metadata:
- repository: semcod/planfile
- planfile_id: semcod/planfile:d2a989f6ada2f6b2:PLF-17304
CONTINUATION-SCOPED-INBOUND553-TIMEOUT-20261011
User continuation via Willman encountered a reproducible transport stall on current published SDKf0a61b05006c948150955d2a449ad2202a25048e/0.1.126. Existing PLF113 providerfailure handling and PLF13804 CAS source work are delivered; this is a fresh bounded diagnostic request, not a competing implementation. ParentWillmanPLF17303/Issue193. Exact configuredcommand in subactor/onedev-agent: planfile sync github --direction from --ticket PLF-17307. It prints Connected and Found1ticket, then provides no closureprojection/result within45s realWillman taskba2f1e91392b44a19cb61fd108e2bead; one retry also exceeded120s and root terminated only its exact owned CLI PID after verifying argv/cwd/startTicks/stdout target. Actual remoteIssue553 isclosed at20:08, localticket remainsopen. NativeSDK/runtime canary itself ishealthy888PASS/160subtests and is not blocked by this bookkeeping transport. Remoteclosure preserved; outbox retained, no repeated outboundreopen or sourcechanges. Cause is not yet localized; do not infer credential/provider/code failure from silence alone.
Acceptance: owning native diagnostic/allocation/currentsourceadmission and independentvalidation. Identify exact blocked stage, per-ticket provider pagination/cache/coalescing and timeout behavior; bind deadlines to monotonic operation lifetime, not fileactivity; verify --ticket selects bounded linkedremoteintake. Timeouts must return explicit nonzero/structuredpartial effects, preserve localCAS/currentleases/sourceidentity, previouscomplete artifacts and remote terminalstate; no success claims or globalimport/forcedstatus bypass. Add intercepted slow/hanging provider and normal exactticket closure controls through publiccommand before independent current-head/base CI. Source fix/runtimeactivation require their ownprotectedscope; this queued request grantsnone.
Evidence: Willman .subactor/receipts/autonomy-20261011/continuation-pr227-actual-publication-queue.json, continuation-pr227-actual-transports.json, continuation-pr227-inbound553-retry.log, continuation-pr227-inbound553-owned-process.json; retained .subactor/recovery/autonomy-20261011/pending-onedev553-inbound.json.
CONTINUATION-OFFLINE-DEPENDENCY-REGRESSIONS-20261011T2100
Bounded diagnosis executed by real Willman task 851a5d86b31e4f76baa7d36c1c7fe436 against frozen published Planfile f0a61b0. Offline SDK/GitHub adapter fixtures confirm sync_from_external calls _fetch_external_tickets before normalizing/resolving ticket_ids; _fetch_external_tickets calls backend.list_tickets, and GitHubBackend._list_tickets enumerates repo.get_issues(state='all'). Thus --ticket PLF-17307 restricts local effects, not provider reads. This is a verified scalability defect and possible explanation for the prior stall, not proof of its live stack/cause or a credential diagnosis.
57 tests PASS, 1 strict XFAIL retains the failing desired contract, one existing PyGithub deprecation warning (2.88s). New controls prove selected closure updates only its mapped local ticket, preserves the unrelated ticket, direct backend.get_ticket('553') uses one exact get_issue, and unrelated list failure preserves both local records while retaining the provider error cause. Desired-contract test demonstrates that selected sync cannot finish if unrelated enumeration is unavailable, although the direct issue read is available. No real API calls in the regression suite, no live inbound retry or remote closure overwrite, no source applied.
Next: admitted owner implements exact mapped inbound retrieval before enumeration where supported, preserving repository identity, unambiguous mappings, remote labels, sprint routing, SDK CAS/live-execution guards, unmanaged/unmapped import semantics, provider failures/cooldown, and a bounded total read deadline with stage diagnostics. Never treat _resolve_selected_remote_ids mixed local/remote candidates as arbitrary API issue IDs. Validate with existing selected/custom-sprint/CAS/live-lease tests and exact/unmapped/ambiguous/error/timeout/PR-negative controls before independent publication. Preserve OneDev553 remotely CLOSED; import only its authoritative closure after the adapter fix.
Reproducer: owned fixture .subactor/cache/continuation2100-dependencies/planfile/tests/test_continuation256_offline.py, SHA d91257583837051fec8f3902277f09b8b58c8396ca221774fb9e1711350f9351; receipts continuation2100-planfile-tests.{log,xml}. No editing grant inferred from this diagnostic.
CONTINUATION-EXACT-INBOUND-RECOVERY-20261011T2123
Latest accepted continuation: continue planned tasks using Willman, verify appropriate tests and report progress. Existing Planfile PLF17304/Issue256 remains the source owner; primary Willman native197 is still unadmitted and no competing source writer was allocated.
Operational recovery of OneDev PLF17307/GitHub subactor/onedev-agent#553 used an exact read adapter over the actual configured published GitHubBackend and public sync_from_external SDK0.1.126/sourcef0a61b0. It calls get_ticket(553), validates issue ID/key/URL/repository and delegates fresh numeric repository identity; it does not enumerate repository issues or change producer code, policy, credentials or runtime. Standard SDK CAS/revision, routing and active execution guards are retained. Actual Willman dry task8cc9d82f was DONE with unchanged local record and fresh closed/completed provider packet. Apply task02837cd7 performed the local closure within35s budget, then failed our post-write assertion expecting absent execution. Published SDK correctly created a finished done execution without assignee/lease. No repeated mutation was attempted; original failure retained. Initial independent readback used an unsupported gh issue JSON field and failed; corrected supported REST readback taskcb7598f4 is DONE and independently verifies localdone, finished execution, no owner/lease, fresh remoteclosed/completed and read-only unchanged record.
Tests task808b53ac DONE:78PASS/2.60s (original71PASS/4.60s retained). Covers YAML/sharded exact selection, stale/concurrent CAS, unrelated concurrent preservation, live execution rejection, wrong provider bindings, labels, failure-cause preservation and valid/invalid terminal execution readback. AdapterSHA7994fa84d8e0869d3f3ca4547c2f1408e927d68eee5e54fff4aac3d1069b483a. OneDev553 inbound outbox now resolved with independent receipt; previous45/120s transport failures and partial-write failure retained. No outbound553 reopening.
This is a bounded operational recovery, not the producer CLI fix: current SDK ticket_ids still filters after list_tickets; original broad CLI stall's exact live cause remains unproven. Source owner must implement bounded exact scoped reads on a fresh admitted native ticket, retain these SDK guards, run native producer tests and protected independent review. No root source commit/push/PR/merge/deployment or autonomy/live campaign activation. Receipts: Willman .subactor/receipts/autonomy-20261011/continuation2123-*.json; managed primary checkpoint18 retains native197 HEAD4f2ffbd, exact plan/slice/snapshot, leaseNULL and AC01 wait.
CONTINUATION-PLANFILE-EXACT-FETCH-CANDIDATE-20261011T2144
Accepted user continuation: resume existing planned tasks through Willman, run appropriate tests, report progress. Existing PLF-17304/Issue256 remains the owner. Primary Willman native197 remains unchanged and unadmitted; no competing implementation writer was allocated.
Prepared a three-file UNAPPLIED proposal against published Planfile f0a61b0. Before: sync_from_external(ticket_ids=["PLF-001"]) lists all GitHub issues, so an unavailable unrelated page blocks an exact issue import. After: resolve selected remote IDs before fetch and use GitHubBackend.list_selected_tickets to read only canonical issue numbers through get_ticket. The existing provider identity proof, SDK revision/snapshot CAS, terminal execution safeguards, sprint scope and labels remain in effect. Selected embedded references are checked against the verified repository; returned ID/key/URL must agree, PRs remain rejected. Unscoped sync and legacy providers without the capability retain their list contract. This does not prove the exact cause of the earlier live45/120s CLI stall, nor change deployed SDK/runtime.
Actual Willman task d4eabc1b qualified baseline negative controls: 8 desired exact-read checks FAIL, 4 compatibility controls PASS (intentional regression detection). Candidate task 7e84bfb3: 388 PASS in12.53s across every tests/test_sync*.py suite, including60 new YAML/sharded cases. Controls cover direct local/numeric/hash/qualified selectors, no enumeration, duplicate reads, fetch failure before writes, selected/unrelated concurrency, stale CAS, live leases, dryrun, labels, wrong packets/references, repository case, fresh numeric proof and identity conflict, custom/unknown sprints and existing list compatibility. Earlier380PASS and harness failure05f00829 (new test used nonexistent move API) are retained; corrected test uses actual Store.move_ticket, candidate source unchanged.
Patch SHA256 8d7f716ee0e043e3f54fe306231e9fcdc8caafd066005f2c4625ef7f07722e88; git apply --check PASS on unchanged current f0a; all candidate Python files parse. Artifact: /home/tom/github/paxlet-com/willman/.subactor/receipts/autonomy-20261011/continuation2144-planfile256-exact-fetch.patch. Qualification: continuation2144-planfile256-qualification.json in the same directory. No source edits to target, new commits, pushes, PR, merge or runtime deployment; isolated ignored snapshots only. Local tests are not protected CI or trusted review.
Next action: existing owner obtains native allocation and current fenced editing admission, reapplies/revalidates on exact current base, runs full required native governance/platform gates, then requests independent Validator approval and protected merge. Preserve completed OneDev553 and its independently verified local terminal execution; no repeated import or outbound reopening.
Activity
tom-sapletta-com commentedon Oct 28, 2026
CONTINUATION-CURRENT-BASE253-20261028
Resumed this existing Planfile request after the same scoped inbound stall occurred during Heyasa autonomy repair. No duplicate intake issue was created.
The existing unapplied patch
8d7f716ee0e043e3f54fe306231e9fcdc8caafd066005f2c4625ef7f07722e88still applies cleanly to current base4adc9a21c3ee03fd11fa4d83ecdccd8dde701147. A current baseline desired-contract case fails because selected issue intake calls unrelated enumeration (sync_inbound_fetch_failed).Fresh external-fixture qualification of the unchanged proposal passes 667 tests across all26current
test_sync*.pyfiles, including its60selected-provider cases; no failures, errors or skips (83.67s). The60proposal cases also passed separately in22.76s. This retains current compatibility beyond the historical388-test result. Disk-backed attempts were bounded/interrupted under journal pressure; the completed results used owned tmpfs data fixtures. No test guard, candidate source, required CI or production data was changed for that mitigation.Source admission remains blocked. Managed allocation reserved native ticket-253, scoped to
planfile/sync/operations.py,planfile/sync/github.pyandtests/test_sync_selected_provider_reads.py. Work-start reports a disjoint application candidate, but the deployed authoritative ChangeLeaseStore refuses main acquisition withchange_lease_resource_already_claimed. Its current main owner is native251 / PR311 at7ca251a8d32f7e54a77e0b1aa9b07dae0c67d127, phase dispatching, revision5/fence318. The repository/target-wide publication lease serializes source writers despite disjoint files. No foreign ownership was taken over or target/namespace changed to avoid the refusal.Native253 is BLOCKED with bounded intent, controller request/refusal, exact owner observation, external snapshot and continuity checkpoint. The candidate remains unapplied: no delivery branch, source commit, push, PR, merge, protected CI or deployment for253. These functional results grant none of that authority.
Next action: resume after legitimate main release or a declared admitted disjoint route; reobserve owner/base/scope, create the canonical delivery worktree and current fenced writer, apply/revalidate the existing proposal, then run native governance and all required exact-head/current-base verification before independent publication. Keep the remaining total-deadline/transport requirements of #256 explicit; do not close the whole issue from this slice.
tom-sapletta-com commentedon Oct 28, 2026
Native253 source admission resumed after the existing owner legitimately released main revision7/fence318. The correction is now committed as
ee90ee103d39f77fd0bc7d7ebe36d510b353cc86and published in PR #312, using its own controller lease and canonical delivery worktree.Actual admitted-worktree validation passes:667affected tests with Testless coverage, Ruff and native governance. No protected check or runtime dependency changed. The head is frozen; required GitHub checks and OneDev job
8b88f47083bf9bd017d40f5f(base8e2b186, profile17cf68d) are pending before independent Validator publication.This updates the earlier blocked/unapplied observation. It is source/PR delivery, not approval, merge or runtime deployment. Remaining total-operation deadline/transport work stays in #256.
tom-sapletta-com commentedon Oct 28, 2026
PR #312 was independently merged at 959224a after protected verification. An additional SDK probe then confirmed a missed numeric-local-ID collision: selecting a local numeric ID could also update an unrelated GitHub issue. Native ticket-254 corrects the selection precedence and rejects conflicting mappings/state lookup failures before issue reads.
Correction: #314 at 116dc03. Six regression cases failed before correction; all 683 affected sync tests now pass, including 76 selected-read cases on YAML and sharded storage. Ruff and native governance pass. The protected OneDev job f15d356d52cb30441ce1e162 is running against main 959224a; independent publication remains required. No runtime deployment is claimed. This is a partial delivery under existing #256, which retains the total-operation deadline/transport requirements.
tom-sapletta-com commentedon Oct 28, 2026
Source delivery confirmed: new-project PR546 merged02721a30ea349a89e6f1be452948322edb2bc251; Planfile PR312 merged959224a9cc95fe0a73a0d26b285c6b17b7c55561; health PR315 mergedbe3b0af32f7c7f82b0665632c0c7ccdf66039502; selector PR314 mergedff2193f8865ee2f83b0171d1cd0e44dd0a8a225f. Each publication has actual exact-head trusted Validator approval. Final PR314 verification passed 1911 tests / 12 skips / 2 warnings on EACH Python3.10,3.11,3.13, plus all eight required checks. Own writer resources are released; this is source delivery, not runtime promotion.
PR314 fixes the reproduced local numeric selector collision without broadening remote issue reads. PR315 preserves safe-loader diagnostics and deadlines while avoiding false healthy results. The broader transport/deadline portion of #256 remains open; do not close it merely because these slices merged. Native health tracking PLF-20016 is done with the real terminal receipt. Central PLF-20347 was refreshed by guarded CAS and readback at17:52:09Z; its runtime dependency remains waiting_dependency. Existing #313 retains the old API evidence-append500/index-liveness adoption gap.