[BUG] Exact-selector race VM stops in physical input observation #64

Open
opened 2026-09-19 11:03:55 +02:00 by erikinkinen · 2 comments
Owner

Summary

An original catalog run fails the existing exact-selector race scenario in the physical QMP input observer. No authority bypass or guest cause is established.

Reproduction

Original CI 1670 runs tests/scenarios/appliance-disk-image-exact-selector-race-negative.toml through the maintained scenario runner. Worker 9 reports physical QMP input injector failed with status 1. The same scenario explicitly passes in original CI 1669; this is not an unchanged workflow rerun.

Expected and actual behavior

Expected: the existing input and exact-selector checks complete with every required marker. Actual: the physical injector exits 1 after repeated screenshot records with no trusted text. The available diagnostic does not identify its underlying exception. Do not infer guest consent, successful redirection, or a common cause with other timeouts.

Environment and identity

Integration 8b1c037aed2503e1f2a4b17c8a8be0666d62a305, original CI 1670/API 9928, integration job 21301 attempt 1. Release scenario, development runtime policy, one virtual CPU, physical selector-race input and original 120-second QEMU ceiling. Exact failed image/firmware bytes are not recovered from the logs. No introducing revision is established.

Evidence and prior checks

Both original workflows have six complete retained logs totaling 27,557,577 bytes without warning candidates. Rust and Markdown succeed. CI 1669 reports 485/486 scenario passes; CI 1670 reports 484/486. Both also retain the ext4 quota timeout. No incompatible shell-layout refusal remains. Later native-image probes are not reached. No deadlines, source/effect/frame gates or workflows were changed to classify this observation.

Original CI observation — 19 September 2026: CI 1671 finishes with 471 passes/16 failures and CI 1672 with 465 passes/22 failures across all 487 scenarios at c479813423fb71c606fa30e5cf7d4dfb16f557d0. Every failed scenario is an ext3/ext4 positive filesystem case reporting error: qemu timed out after 120s, with scenario status 1. The existing filesystem deadline report retains the expanded matrix; quota issue 18 applies to the quota failure in 1672. All six complete logs are hashed (41,592,478 bytes), with zero warning candidates. Rust, Markdown and the full helper step pass. The zero-realm startup and exact-selector race scenarios explicitly pass in both runs. No common cause, source regression boundary or performance diagnosis is inferred; all actual deadlines and failure statuses remain unchanged, and neither workflow was rerun. The original failed observer exception remains unestablished; paired later passes alone do not identify or repair its cause.

Tracking

Integration PR 12, ext4 quota issue 18. Recover the original observer exception and inner evidence before attributing cause. Preserve the real consent and identity checks.

## Summary An original catalog run fails the existing exact-selector race scenario in the physical QMP input observer. No authority bypass or guest cause is established. ## Reproduction Original [CI 1670](https://git.erikinkinen.fi/erix/integration/actions/runs/1670) runs `tests/scenarios/appliance-disk-image-exact-selector-race-negative.toml` through the maintained scenario runner. Worker 9 reports `physical QMP input injector failed with status 1`. The same scenario explicitly passes in original [CI 1669](https://git.erikinkinen.fi/erix/integration/actions/runs/1669); this is not an unchanged workflow rerun. ## Expected and actual behavior Expected: the existing input and exact-selector checks complete with every required marker. Actual: the physical injector exits 1 after repeated screenshot records with no trusted text. The available diagnostic does not identify its underlying exception. Do not infer guest consent, successful redirection, or a common cause with other timeouts. ## Environment and identity Integration `8b1c037aed2503e1f2a4b17c8a8be0666d62a305`, original CI 1670/API 9928, integration job 21301 attempt 1. Release scenario, development runtime policy, one virtual CPU, physical selector-race input and original 120-second QEMU ceiling. Exact failed image/firmware bytes are not recovered from the logs. No introducing revision is established. ## Evidence and prior checks Both original workflows have six complete retained logs totaling 27,557,577 bytes without warning candidates. Rust and Markdown succeed. CI 1669 reports 485/486 scenario passes; CI 1670 reports 484/486. Both also retain the ext4 quota timeout. No incompatible shell-layout refusal remains. Later native-image probes are not reached. No deadlines, source/effect/frame gates or workflows were changed to classify this observation. Original CI observation — 19 September 2026: [CI 1671](https://git.erikinkinen.fi/erix/integration/actions/runs/1671) finishes with 471 passes/16 failures and [CI 1672](https://git.erikinkinen.fi/erix/integration/actions/runs/1672) with 465 passes/22 failures across all 487 scenarios at `c479813423fb71c606fa30e5cf7d4dfb16f557d0`. Every failed scenario is an ext3/ext4 positive filesystem case reporting `error: qemu timed out after 120s`, with scenario status 1. The existing [filesystem deadline report](https://git.erikinkinen.fi/erix/integration/issues/20) retains the expanded matrix; [quota issue 18](https://git.erikinkinen.fi/erix/integration/issues/18) applies to the quota failure in 1672. All six complete logs are hashed (41,592,478 bytes), with zero warning candidates. Rust, Markdown and the full helper step pass. The zero-realm startup and exact-selector race scenarios explicitly pass in both runs. No common cause, source regression boundary or performance diagnosis is inferred; all actual deadlines and failure statuses remain unchanged, and neither workflow was rerun. The original failed observer exception remains unestablished; paired later passes alone do not identify or repair its cause. ## Tracking [Integration PR 12](https://git.erikinkinen.fi/erix/integration/pulls/12), [ext4 quota issue 18](https://git.erikinkinen.fi/erix/integration/issues/18). Recover the original observer exception and inner evidence before attributing cause. Preserve the real consent and identity checks.
Author
Owner

CI 2732 records a
new failure in the same selector-race scenario at Integration
0e93075a477efa06a39007c206ebf5bc8fd023be (VM job 47544, attempt 1, task
34524, worker 9).

The retained live prefix contains RACESETUP0, the next directory-creation
consent dialog, exsh: command expansion failed, a later shell prompt, and then
the unchanged 45-second serial progress watchdog at 12:35:58 UTC on 16 October.
QEMU is terminated and image-profile preflight fails. This differs from the
original blank-screenshot physical-input failure reported here; a shared cause
has not been established.

The task was still running before and after this single progress read. The
complete terminal log and inner VM artifacts remain necessary before assigning
a final matrix outcome or cause. The original consent, selector and deadline
checks remain in place.

[CI 2732](https://git.erikinkinen.fi/erix/integration/actions/runs/2732) records a new failure in the same selector-race scenario at Integration `0e93075a477efa06a39007c206ebf5bc8fd023be` (VM job 47544, attempt 1, task 34524, worker 9). The retained live prefix contains `RACESETUP0`, the next directory-creation consent dialog, `exsh: command expansion failed`, a later shell prompt, and then the unchanged 45-second serial progress watchdog at 12:35:58 UTC on 16 October. QEMU is terminated and image-profile preflight fails. This differs from the original blank-screenshot physical-input failure reported here; a shared cause has not been established. The task was still running before and after this single progress read. The complete terminal log and inner VM artifacts remain necessary before assigning a final matrix outcome or cause. The original consent, selector and deadline checks remain in place.
Author
Owner

Reproduced in CI run 2822, Integration f06bf722. The exact-selector-race scenario reaches ERIX_ROOTD:PHYSICAL_INPUT:EVIDENCE_READY and prints RACESETUP0, then hits the 45-second serial-progress watchdog at 15:17:06 UTC on 17 October. The prompt stops at the following slash. The point of input loss still needs isolating.

Reproduced in [CI run 2822](https://git.erikinkinen.fi/erix/integration/actions/runs/2822), Integration `f06bf722`. The exact-selector-race scenario reaches `ERIX_ROOTD:PHYSICAL_INPUT:EVIDENCE_READY` and prints `RACESETUP0`, then hits the 45-second serial-progress watchdog at 15:17:06 UTC on 17 October. The prompt stops at the following slash. The point of input loss still needs isolating.
erikinkinen added this to the Alpha 1 milestone 2026-10-30 09:34:15 +01:00
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
erix/integration#64
No description provided.