[BUG] Named idle-time-resolution VM reaches its original QEMU deadline #61

New issue

Open

opened 2026-09-17 20:01:05 +02:00 by erikinkinen · 0 comments

erikinkinen commented

2026-09-17 20:01:05 +02:00

Owner

Summary

The existing named idle-time-resolution VM scenario reaches its 60-second QEMU deadline in original Integration PR CI 1652. It passes in the paired original push run 1651 at the same source revision. The cause and a deterministic isolated reproduction remain unestablished; the failed execution remains a catalog regression gate.

Reproduction

Observed on 17 September 2026 in original PR CI 1652, worker 3, at Integration fa86d1b64f8e1299c0ee0c886e6f1b35d58ea15b:

tests/scenarios/subsystem-named-idle-time-resolution-positive.toml

The ordinary full catalog reports qemu timed out after 60s. Original push CI 1651 passes this scenario without rerunning the failed input. Preserve the original catalog, scenario, component selections, emulator configuration and 60-second limit when constructing the next bounded diagnostic. An isolated reproduction is not yet established.

Expected and actual behavior

Expected: target-scoped timed resolution wakes the idle named server, the client succeeds, all declared readiness/process-event markers appear, and the guest exits with the declared 0x10 result before the original deadline.

Actual: QEMU is terminated at the deadline. The retained serial dump contains the NAMED_READY, TIMED_READY and TIME_READY stages, followed by continued paging and scheduling diagnostics. The required zero-status process event and standalone final ROOTD:READY line are absent. Service-prefixed ROOTD:READY lines are not evidence of that final marker.

CI 1652 records 406 passes and 80 failures out of 486: 78 known layout refusals (#57), the ext4 quota timeout (#18), and this scenario. CI 1651 records 407 passes and 79 failures: the layout refusals and quota timeout. Neither full workflow passes.

Environment and identity

Exact Integration head and original run identities are above; no replacement run is used.
Full-system x86-64 VM, release profile, one guest CPU, runtime configuration tests/runtime-configs/runtime.dev.vm.toml.
Root orchestration mode time; named/timed and supporting service smoke profiles; original QEMU limit 60 seconds.
The serial tail was extracted only from worker 3 between this scenario and the next worker-owned scenario. Log replay timestamps do not measure guest event timing.
Complete original image, firmware and toolchain byte equivalence has not been established. The paired pass is not a broader last-known-working revision.
No new capability route or authority grant is introduced by this report, and no guest authority violation is asserted.

Evidence and prior checks

All six original terminal logs for CI 1651/1652 are complete and hash-bound, totaling 27,230,062 bytes with zero warning candidates. Rust and Markdown jobs pass in both runs; each full VM job fails with the exact counts above. The failure set and timeout are independently identified from the worker log and catalog summary. A shared timing symptom does not establish a common cause with #18 or other storage timeouts.

No unchanged rerun, deadline increase, marker relaxation or speculative source change is claimed as a correction.

Tracking

Owning validation: #1 and #12. Related retained failures: #18 and #57. Keep this report open until original source and scenario identities support a bounded reproduction, the cause is identified, and a correction passes focused and full-catalog checks with unchanged timing, authority and cleanup requirements.

## Summary The existing named idle-time-resolution VM scenario reaches its 60-second QEMU deadline in original Integration PR CI 1652. It passes in the paired original push run 1651 at the same source revision. The cause and a deterministic isolated reproduction remain unestablished; the failed execution remains a catalog regression gate. ## Reproduction Observed on 17 September 2026 in [original PR CI 1652](https://git.erikinkinen.fi/erix/integration/actions/runs/1652), worker 3, at Integration `fa86d1b64f8e1299c0ee0c886e6f1b35d58ea15b`: `tests/scenarios/subsystem-named-idle-time-resolution-positive.toml` The ordinary full catalog reports `qemu timed out after 60s`. [Original push CI 1651](https://git.erikinkinen.fi/erix/integration/actions/runs/1651) passes this scenario without rerunning the failed input. Preserve the original catalog, scenario, component selections, emulator configuration and 60-second limit when constructing the next bounded diagnostic. An isolated reproduction is not yet established. ## Expected and actual behavior Expected: target-scoped timed resolution wakes the idle named server, the client succeeds, all declared readiness/process-event markers appear, and the guest exits with the declared `0x10` result before the original deadline. Actual: QEMU is terminated at the deadline. The retained serial dump contains the NAMED_READY, TIMED_READY and TIME_READY stages, followed by continued paging and scheduling diagnostics. The required zero-status process event and standalone final ROOTD:READY line are absent. Service-prefixed ROOTD:READY lines are not evidence of that final marker. CI 1652 records 406 passes and 80 failures out of 486: 78 known layout refusals (#57), the ext4 quota timeout (#18), and this scenario. CI 1651 records 407 passes and 79 failures: the layout refusals and quota timeout. Neither full workflow passes. ## Environment and identity - Exact Integration head and original run identities are above; no replacement run is used. - Full-system x86-64 VM, release profile, one guest CPU, runtime configuration `tests/runtime-configs/runtime.dev.vm.toml`. - Root orchestration mode `time`; named/timed and supporting service smoke profiles; original QEMU limit 60 seconds. - The serial tail was extracted only from worker 3 between this scenario and the next worker-owned scenario. Log replay timestamps do not measure guest event timing. - Complete original image, firmware and toolchain byte equivalence has not been established. The paired pass is not a broader last-known-working revision. - No new capability route or authority grant is introduced by this report, and no guest authority violation is asserted. ## Evidence and prior checks All six original terminal logs for CI 1651/1652 are complete and hash-bound, totaling 27,230,062 bytes with zero warning candidates. Rust and Markdown jobs pass in both runs; each full VM job fails with the exact counts above. The failure set and timeout are independently identified from the worker log and catalog summary. A shared timing symptom does not establish a common cause with #18 or other storage timeouts. No unchanged rerun, deadline increase, marker relaxation or speculative source change is claimed as a correction. ## Tracking Owning validation: #1 and #12. Related retained failures: #18 and #57. Keep this report open until original source and scenario identities support a bounded reproduction, the cause is identified, and a correction passes focused and full-catalog checks with unchanged timing, authority and cleanup requirements.

erikinkinen added the

bug

phase-6

performance

labels

2026-09-17 20:01:05 +02:00

erikinkinen referenced this issue

2026-09-17 20:01:06 +02:00

[FEATURE] Freeze the complete source, build, ABI and self-hosting inventory #1

erikinkinen referenced this issue

2026-09-17 20:01:06 +02:00

WIP: Coordinate native authority, shell workspace and source transport #12