[BUG] Namespace cancellation scenario stops during extended VFS checks #176

Open
opened 2026-10-21 01:44:47 +02:00 by erikinkinen · 2 comments
Owner

Summary

The namespace-cancellation VM scenario launches successfully but stops during the
subsequent extended VFS checks. Its namespace-cancellation success marker is
present; the full scenario does not reach final readiness.

Reproduction

With Integration 8baabcb32f54259cd8ebc3129289a8d798672de7, prepare the selected
component checkouts and offline build inputs, then run:

bash scripts/run-scenario.sh tests/scenarios/subsystem-vfs-namespace-cancellation-positive.toml components.toml

This was observed once with retained build caches. A clean-build reproduction
and the exact failing VFS operation have not yet been established.

Expected and actual behavior

The original scenario should return QEMU status 33 and reach all required
readiness markers. Instead, QEMU returns 41 and reports
ERIX_KERNEL:ROOTD_SMOKE:SERVICE_FAIL.

The complete serial log records this order:

  1. ERIX_ROOTD:READY:VFS_NAMESPACE_CANCELLATION
  2. ERIX_ROOTD:PROGRESS:VFS:TMP_IO_DONE
  3. ERIX_KERNEL:ROOTD_SMOKE:SERVICE_FAIL

The selected memory pipeline runs its extended VFS checks after /tmp I/O.
That identifies the failing phase interval, not a particular request or cause.

Environment and identity

Linux x86-64 host, Rust 1.97.1, offline Cargo with one build job, release image,
one QEMU CPU and the original 120-second guest deadline. The retained image is
c16358b29a5ddaf8a30e3b249e49930ef3143ac9c9f1465919ecde154eef87fb.
Its Rootd metadata identifies e45718b27d18d74d015b584445ca1332fb1c6b04 and the
namespace-cancellation feature. No last working revision is known.

Evidence and prior checks

The original serial output, image, command result and process cleanup records
are retained. Image compilation, signing and verification completed before QEMU
started; this is a later failure than the packaging defect in
#172.
The VM process retired cleanly. Source and artifact provenance are being checked
separately; a checkout revision alone is not proof of compiler consumption.

Tracking

Related work: PR 12.

  • Retain the complete failed guest output and its exact image.
  • Identify the original failing VFS operation and result.
  • Correct the cause while preserving namespace custody and cleanup rules.
  • Cover the failing operation with a regression test.
  • Pass the complete original VM scenario with all required markers.
## Summary The namespace-cancellation VM scenario launches successfully but stops during the subsequent extended VFS checks. Its namespace-cancellation success marker is present; the full scenario does not reach final readiness. ## Reproduction With Integration `8baabcb32f54259cd8ebc3129289a8d798672de7`, prepare the selected component checkouts and offline build inputs, then run: ```sh bash scripts/run-scenario.sh tests/scenarios/subsystem-vfs-namespace-cancellation-positive.toml components.toml ``` This was observed once with retained build caches. A clean-build reproduction and the exact failing VFS operation have not yet been established. ## Expected and actual behavior The original scenario should return QEMU status 33 and reach all required readiness markers. Instead, QEMU returns 41 and reports `ERIX_KERNEL:ROOTD_SMOKE:SERVICE_FAIL`. The complete serial log records this order: 1. `ERIX_ROOTD:READY:VFS_NAMESPACE_CANCELLATION` 2. `ERIX_ROOTD:PROGRESS:VFS:TMP_IO_DONE` 3. `ERIX_KERNEL:ROOTD_SMOKE:SERVICE_FAIL` The selected memory pipeline runs its extended VFS checks after `/tmp` I/O. That identifies the failing phase interval, not a particular request or cause. ## Environment and identity Linux x86-64 host, Rust 1.97.1, offline Cargo with one build job, release image, one QEMU CPU and the original 120-second guest deadline. The retained image is `c16358b29a5ddaf8a30e3b249e49930ef3143ac9c9f1465919ecde154eef87fb`. Its Rootd metadata identifies `e45718b27d18d74d015b584445ca1332fb1c6b04` and the namespace-cancellation feature. No last working revision is known. ## Evidence and prior checks The original serial output, image, command result and process cleanup records are retained. Image compilation, signing and verification completed before QEMU started; this is a later failure than the packaging defect in [#172](https://git.erikinkinen.fi/erix/integration/issues/172). The VM process retired cleanly. Source and artifact provenance are being checked separately; a checkout revision alone is not proof of compiler consumption. ## Tracking Related work: [PR 12](https://git.erikinkinen.fi/erix/integration/pulls/12). - [x] Retain the complete failed guest output and its exact image. - [ ] Identify the original failing VFS operation and result. - [ ] Correct the cause while preserving namespace custody and cleanup rules. - [ ] Cover the failing operation with a regression test. - [x] Pass the complete original VM scenario with all required markers.
Author
Owner

A separate hosted reproduction found that an expected application refusal was repeated in the final disposal summary, overriding the caller’s successful semantic result. The coordinated LibVfs feb949af / Rootd b4c7c4b7 change handles that case while preserving strict default policy, unexpected semantic errors and independent cleanup failures. Vfsd 00d6a73c supplies the fresh complete producer handoff.

The component checks and hosted handoff pass. This does not establish the original guest’s exact failure. The next image will use the coordinated source graph and a freshly generated revision vector before the guest is run again.

A separate hosted reproduction found that an expected application refusal was repeated in the final disposal summary, overriding the caller’s successful semantic result. The coordinated LibVfs feb949af / Rootd b4c7c4b7 change handles that case while preserving strict default policy, unexpected semantic errors and independent cleanup failures. Vfsd 00d6a73c supplies the fresh complete producer handoff. The component checks and hosted handoff pass. This does not establish the original guest’s exact failure. The next image will use the coordinated source graph and a freshly generated revision vector before the guest is run again.
Author
Owner

The complete original scenario now passes with the published Rootd correction and LibVfs correction. The build creates a fresh 31-object registry, signs and verifies the image, and QEMU exits 33 within the unchanged 120-second guest deadline. All required readiness markers and native cleanup checks pass.

Checked the full-scenario criterion. Keeping the investigation open: the original status-41 log still does not identify the exact failing request and result. The hosted refusal-policy controls establish that separate defect; they do not prove the original guest failure cause. All original failure records remain retained.

The complete original scenario now passes with the published [Rootd correction](https://git.erikinkinen.fi/erix/rootd/commit/b4c7c4b7) and [LibVfs correction](https://git.erikinkinen.fi/erix/lib-vfs/commit/feb949af). The build creates a fresh 31-object registry, signs and verifies the image, and QEMU exits 33 within the unchanged 120-second guest deadline. All required readiness markers and native cleanup checks pass. Checked the full-scenario criterion. Keeping the investigation open: the original status-41 log still does not identify the exact failing request and result. The hosted refusal-policy controls establish that separate defect; they do not prove the original guest failure cause. All original failure records remain retained.
erikinkinen added this to the Alpha 1 milestone 2026-10-30 09:34:20 +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#176
No description provided.