[BUG] Release COM1 directory CI probe fails after catalog and earlier directory checks pass #36

New issue

Open

opened 2026-09-12 23:50:32 +02:00 by erikinkinen · 0 comments

erikinkinen commented

2026-09-12 23:50:32 +02:00

Owner

Summary

The positive release-image directory and current-working-directory probe over COM1 failed in Integration push run 1581 at 9e8e2917de855d6bf42285c0f0cfabb116790b26. All 486 catalog scenarios and the preceding development physical, development COM1 and release physical directory probes passed. The retained diagnostic reports serial-input failure together with composed VM-command status 125; it does not identify the failed command or establish a cause.

The same release COM1 gate and sanitized status tuple were observed again in push run 1583 at 4658ee9fab995eab1812f219971cffd340e1eb70. This is a repeated observed outcome; it does not establish a shared cause.

Reproduction

The observed workflow step was Validate release directory tools and cwd through COM1:

./tests/native-directory-vm.sh serial out/appliance/erix-release.img

This is the command from the original workflow at the reported revision, using that workflow's release image. The native-directory profile checks 14 commands, their exact status/output, consent and denial behavior, current-directory preservation and fresh prompts. This report records retained CI observations; no replay was performed and the failure frequency is unknown.

Expected and actual behavior

Expected: the positive release COM1 directory probe completes all 14 steps and returns 0.

Actual: the attempt began at 2026-09-12 21:24:45.6637880 UTC. At 21:25:03.4311365 UTC, the sanitized diagnostic was:

vm diagnostic: stage=serial-input runner_exit=1 vm_exit=125 expected_vm_exit=0 physical_input_exit=unknown serial_input_exit=1 physical_cleanup_requested=0

The wrapper then reported failure and returned 1. The main job became terminal failure; Rust and Markdown jobs succeeded. The original runner checks the waited serial-input failure before its composed VM-status branches. vm_exit=125 belongs to the composed watchdog/timeout/QEMU command, not an independently recovered raw QEMU exit or proof of one particular watchdog branch. The physical-cleanup field does not establish that all cleanup was unnecessary or absent.

Environment and identity

Repository/revision: Integration 9e8e2917de855d6bf42285c0f0cfabb116790b26, push run 1581, main job 19528, attempt 1. The component graph is the original components.toml and cargo-repos.toml at that revision.
Image/route: workflow-built release image, selected through the maintained native-directory-vm.sh wrapper and serial COM1 route. The failing probe image digest was not recovered from the retained job log; no artifact download was made.
Declared runner configuration: x86-64 QEMU, one virtual CPU, 120-second VM limit, 15-second progress stall limit and 10-second command limit. Effective firmware, acceleration and executable byte identities were not independently recovered for this CI attempt.
Native-directory expectations include successful consented enumeration and explicit denied operations; expected command status 1 is limited to steps 5, 6 and 10. A failed cd must preserve the current directory.
No introducing revision or last known working revision for this exact release COM1 observation has been established.

Evidence and prior checks

Integration run 1581 retains the failed workflow outcome. All three terminal logs were captured completely once: 13,247,869 bytes total. The main log is 13,143,195 bytes, SHA-256 b5b7b853dab9e61b31d573e0240a4d21512d7a8ca03bf8fb9fefc6309a0d197c.

One-based raw LF line anchors in that main log are: catalog summary 100214 (passed=486 failed=0 total=486); development physical directory success 100772; development COM1 directory success 100774; release physical directory success 100776; release COM1 start 100777; sanitized diagnostic 100778; wrapper failure 100779. The 486 distinct worker pass lines were checked against the original scenario inventory. The original workflow and relevant wrapper, injector, runner and diagnostic sources were retained and bound to 16 original Git blobs. The API records the main job stopped at 2026-09-12 21:26:33 UTC.

Earlier intentional negative host-fixture diagnostics are separate from this terminal positive-probe failure. The available final job log does not identify which directory command, consent/output check, transport operation or cleanup operation failed. Host contention, a guest fault and a shared cause with other probe failures are not established. No timeout or acceptance gate was changed and no failing probe was rerun.

A later retained observation, Integration run 1583, failed in main job 19542, attempt 1, at 4658ee9fab995eab1812f219971cffd340e1eb70. All 486 distinct original catalog worker scenarios passed, followed by successful development physical, development COM1 and release physical directory probes. The release COM1 attempt began at 2026-09-12 23:07:15.6214714 UTC; at 23:07:33.6888294 UTC it emitted the same sanitized diagnostic quoted above and returned 1. The API terminal time is 23:08:34 UTC. Rust and Markdown jobs succeeded.

All three run-1583 logs were retained completely once, totaling 13,247,856 bytes. Its main log is 13,143,193 bytes, SHA-256 64c09ff389cca65946da4447979dd32c24121210fbfcdd5c86a044d7887d5cd4. One-based raw LF anchors are catalog summary 100214; development physical, development COM1 and release physical directory successes 100772, 100774 and 100776; release COM1 start 100777; diagnostic 100778; wrapper failure 100779. Six earlier host-fixture pass markers are separate from the 486 real catalog worker passes. Sixteen original source blobs were bound to the new revision: fourteen workflow/runner/profile/diagnostic blobs are unchanged from run 1581, while the two catalogs contain the separately published Deviced revision update. That source difference does not establish an introducing change or a failure cause. The failing image digest and detailed injector operation remain unavailable in the retained log.

Tracking

Related: Integration #35, which records the same positive profile family failing on the development physical route at an earlier revision with composed VM status 0. This report concerns the release COM1 route with composed VM status 125; their common cause is unproven. Integration PR #12 coordinates the phase-6 work. No workaround or correction is claimed here.

## Summary The positive release-image directory and current-working-directory probe over COM1 failed in Integration push run 1581 at `9e8e2917de855d6bf42285c0f0cfabb116790b26`. All 486 catalog scenarios and the preceding development physical, development COM1 and release physical directory probes passed. The retained diagnostic reports serial-input failure together with composed VM-command status 125; it does not identify the failed command or establish a cause. The same release COM1 gate and sanitized status tuple were observed again in push run 1583 at `4658ee9fab995eab1812f219971cffd340e1eb70`. This is a repeated observed outcome; it does not establish a shared cause. ## Reproduction The observed workflow step was **Validate release directory tools and cwd through COM1**: ```sh ./tests/native-directory-vm.sh serial out/appliance/erix-release.img ``` This is the command from the original workflow at the reported revision, using that workflow's release image. The native-directory profile checks 14 commands, their exact status/output, consent and denial behavior, current-directory preservation and fresh prompts. This report records retained CI observations; no replay was performed and the failure frequency is unknown. ## Expected and actual behavior Expected: the positive release COM1 directory probe completes all 14 steps and returns 0. Actual: the attempt began at 2026-09-12 21:24:45.6637880 UTC. At 21:25:03.4311365 UTC, the sanitized diagnostic was: ```text vm diagnostic: stage=serial-input runner_exit=1 vm_exit=125 expected_vm_exit=0 physical_input_exit=unknown serial_input_exit=1 physical_cleanup_requested=0 ``` The wrapper then reported failure and returned 1. The main job became terminal failure; Rust and Markdown jobs succeeded. The original runner checks the waited serial-input failure before its composed VM-status branches. `vm_exit=125` belongs to the composed watchdog/timeout/QEMU command, not an independently recovered raw QEMU exit or proof of one particular watchdog branch. The physical-cleanup field does not establish that all cleanup was unnecessary or absent. ## Environment and identity - Repository/revision: Integration `9e8e2917de855d6bf42285c0f0cfabb116790b26`, push run 1581, main job 19528, attempt 1. The component graph is the original `components.toml` and `cargo-repos.toml` at that revision. - Image/route: workflow-built release image, selected through the maintained `native-directory-vm.sh` wrapper and serial COM1 route. The failing probe image digest was not recovered from the retained job log; no artifact download was made. - Declared runner configuration: x86-64 QEMU, one virtual CPU, 120-second VM limit, 15-second progress stall limit and 10-second command limit. Effective firmware, acceleration and executable byte identities were not independently recovered for this CI attempt. - Native-directory expectations include successful consented enumeration and explicit denied operations; expected command status 1 is limited to steps 5, 6 and 10. A failed `cd` must preserve the current directory. - No introducing revision or last known working revision for this exact release COM1 observation has been established. ## Evidence and prior checks [Integration run 1581](https://git.erikinkinen.fi/erix/integration/actions/runs/1581) retains the failed workflow outcome. All three terminal logs were captured completely once: 13,247,869 bytes total. The main log is 13,143,195 bytes, SHA-256 `b5b7b853dab9e61b31d573e0240a4d21512d7a8ca03bf8fb9fefc6309a0d197c`. One-based raw LF line anchors in that main log are: catalog summary 100214 (`passed=486 failed=0 total=486`); development physical directory success 100772; development COM1 directory success 100774; release physical directory success 100776; release COM1 start 100777; sanitized diagnostic 100778; wrapper failure 100779. The 486 distinct worker pass lines were checked against the original scenario inventory. The original workflow and relevant wrapper, injector, runner and diagnostic sources were retained and bound to 16 original Git blobs. The API records the main job stopped at 2026-09-12 21:26:33 UTC. Earlier intentional negative host-fixture diagnostics are separate from this terminal positive-probe failure. The available final job log does not identify which directory command, consent/output check, transport operation or cleanup operation failed. Host contention, a guest fault and a shared cause with other probe failures are not established. No timeout or acceptance gate was changed and no failing probe was rerun. A later retained observation, [Integration run 1583](https://git.erikinkinen.fi/erix/integration/actions/runs/1583), failed in main job 19542, attempt 1, at `4658ee9fab995eab1812f219971cffd340e1eb70`. All 486 distinct original catalog worker scenarios passed, followed by successful development physical, development COM1 and release physical directory probes. The release COM1 attempt began at 2026-09-12 23:07:15.6214714 UTC; at 23:07:33.6888294 UTC it emitted the same sanitized diagnostic quoted above and returned 1. The API terminal time is 23:08:34 UTC. Rust and Markdown jobs succeeded. All three run-1583 logs were retained completely once, totaling 13,247,856 bytes. Its main log is 13,143,193 bytes, SHA-256 `64c09ff389cca65946da4447979dd32c24121210fbfcdd5c86a044d7887d5cd4`. One-based raw LF anchors are catalog summary 100214; development physical, development COM1 and release physical directory successes 100772, 100774 and 100776; release COM1 start 100777; diagnostic 100778; wrapper failure 100779. Six earlier host-fixture pass markers are separate from the 486 real catalog worker passes. Sixteen original source blobs were bound to the new revision: fourteen workflow/runner/profile/diagnostic blobs are unchanged from run 1581, while the two catalogs contain the separately published Deviced revision update. That source difference does not establish an introducing change or a failure cause. The failing image digest and detailed injector operation remain unavailable in the retained log. ## Tracking Related: [Integration #35](https://git.erikinkinen.fi/erix/integration/issues/35), which records the same positive profile family failing on the **development physical** route at an earlier revision with composed VM status 0. This report concerns the **release COM1** route with composed VM status 125; their common cause is unproven. [Integration PR #12](https://git.erikinkinen.fi/erix/integration/pulls/12) coordinates the phase-6 work. No workaround or correction is claimed here.

erikinkinen added the

bug

phase-6

labels

2026-09-12 23:50:32 +02:00

~~erikinkinen referenced this issue 2026-09-12 23:50:33 +02:00~~

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

erikinkinen referenced this issue

2026-09-13 00:07:55 +02:00

[BUG] Release COM1 editor CI probe fails after catalog and earlier editor checks pass #37

erikinkinen referenced this issue

2026-09-13 02:30:25 +02:00