[PR #1155] [MERGED] fix(artifacts): recheck failed DMG detaches before reporting cleanup failure #2689

New issue

Closed

opened 2026-10-17 21:09:05 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-17 21:09:05 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/morluto/rea/pull/1155
Author: @kaoru0822-kitauji
Created: 10/9/2026
Status: ✅ Merged
Merged: 10/9/2026
Merged by: @N0zoM1z0

Base: main ← Head: fix/dmg-cleanup-recheck

📝 Commits (2)

5971c81 fix(artifacts): recheck failed DMG detaches before reporting cleanup failure
9de9925 test(artifacts): share DMG detach fixtures to stay under the duplication gate

📊 Changes

2 files changed (+175 additions, -70 deletions)

View changed files

📝 src/artifacts/NativeDmgArtifactReader.ts (+46 -13)
📝 tests/boundary/providers/native/nativeDmgArtifactReader.test.ts (+129 -57)

📄 Description

Problem

NativeDmgArtifactReader.close() detaches each device once. When a detach fails, it asks hdiutil info once whether the device is still attached, and counts it as a cleanup failure if so. That single check is racy:

On APFS images, detaching the synthesized container also ejects the image disk. hdiutil info can still list the ejected device for a moment.
A volume that was just read can briefly report busy.

In both cases the cleanup actually succeeds or would succeed, but close() throws DMG detach or mount-root cleanup failed and keeps the mount root. Inventory and extraction callers then report the false failure.

This is the intermittent "Real Apple artifact CLI/MCP E2E" failure in verify:macos-bundle (for example on #1070, #1097 and fix/hopper-block-endpoints). Since #1103 added the cause to the message, the job log shows the mechanism:

DMG detach or mount-root cleanup failed for ".../MacFixture.dmg": hdiutil detach failed: {"arguments":["detach","/dev/disk23"],"exitCode":1,"stderr":"hdiutil: detach failed - No such file or directory\n"}

The device was already gone, but it was still counted as attached. The #1103 review raised the same missing recheck ("Recheck failed devices after completing the detach loop"), and it was still unaddressed on main.

Change

After the first detach pass, recheck the remaining devices up to three times, after 250, 500 and 1000 ms:
- drop any device hdiutil info no longer lists;
- retry the detach of any device still listed;
- fail only for devices that stay attached through every recheck.
A successful pass skips the waits entirely. A truly stuck device still fails, with its latest hdiutil diagnostic and the mount root kept, as before.
NativeDmgHost gains an optional delay, so tests exercise the recheck without real waits. The attach-failure cleanup path goes through close() and gets the same behavior.

Verification

npm run test:focused -- tests/boundary/providers/native/nativeDmgArtifactReader.test.ts. Two new cases: a device listed once more after its container's detach ejected it, which matches the CI log; and a device that is busy on its first detach, then detaches within the same close. Both fail on main. The existing APFS, retained-attachment and attach-cleanup cases still pass.
The real-hdiutil test "classifies a real hdiutil verify refusal…" fails on this Mac identically on unmodified main. It's environmental and passes on CI runners.
Real macOS 26.7 check with this branch: 3 HFS+ and 5 APFS hdiutil create images through rea inspect-artifact. All succeeded, and hdiutil info showed no leftover devices.
npm run check:fast

No real Hopper or Ghidra verification was run, because this change does not touch those providers.

🤖 Generated with Claude Code

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/morluto/rea/pull/1155 **Author:** [@kaoru0822-kitauji](https://github.com/kaoru0822-kitauji) **Created:** 10/9/2026 **Status:** ✅ Merged **Merged:** 10/9/2026 **Merged by:** [@N0zoM1z0](https://github.com/N0zoM1z0) **Base:** `main` ← **Head:** `fix/dmg-cleanup-recheck` --- ### 📝 Commits (2) - [`5971c81`](https://github.com/morluto/rea/commit/5971c8165e45549bb42099fc28a0e610aca8807b) fix(artifacts): recheck failed DMG detaches before reporting cleanup failure - [`9de9925`](https://github.com/morluto/rea/commit/9de9925390930b53e78196fa2c9ade4d942ebf8f) test(artifacts): share DMG detach fixtures to stay under the duplication gate ### 📊 Changes **2 files changed** (+175 additions, -70 deletions) <details> <summary>View changed files</summary> 📝 `src/artifacts/NativeDmgArtifactReader.ts` (+46 -13) 📝 `tests/boundary/providers/native/nativeDmgArtifactReader.test.ts` (+129 -57) </details> ### 📄 Description ## Problem `NativeDmgArtifactReader.close()` detaches each device once. When a detach fails, it asks `hdiutil info` once whether the device is still attached, and counts it as a cleanup failure if so. That single check is racy: - On APFS images, detaching the synthesized container also ejects the image disk. `hdiutil info` can still list the ejected device for a moment. - A volume that was just read can briefly report busy. In both cases the cleanup actually succeeds or would succeed, but `close()` throws `DMG detach or mount-root cleanup failed` and keeps the mount root. Inventory and extraction callers then report the false failure. This is the intermittent "Real Apple artifact CLI/MCP E2E" failure in `verify:macos-bundle` (for example on #1070, #1097 and `fix/hopper-block-endpoints`). Since #1103 added the cause to the message, the job log shows the mechanism: ``` DMG detach or mount-root cleanup failed for ".../MacFixture.dmg": hdiutil detach failed: {"arguments":["detach","/dev/disk23"],"exitCode":1,"stderr":"hdiutil: detach failed - No such file or directory\n"} ``` The device was already gone, but it was still counted as attached. The #1103 review raised the same missing recheck ("Recheck failed devices after completing the detach loop"), and it was still unaddressed on `main`. ## Change - After the first detach pass, recheck the remaining devices up to three times, after 250, 500 and 1000 ms: - drop any device `hdiutil info` no longer lists; - retry the detach of any device still listed; - fail only for devices that stay attached through every recheck. - A successful pass skips the waits entirely. A truly stuck device still fails, with its latest `hdiutil` diagnostic and the mount root kept, as before. - `NativeDmgHost` gains an optional `delay`, so tests exercise the recheck without real waits. The attach-failure cleanup path goes through `close()` and gets the same behavior. ## Verification - `npm run test:focused -- tests/boundary/providers/native/nativeDmgArtifactReader.test.ts`. Two new cases: a device listed once more after its container's detach ejected it, which matches the CI log; and a device that is busy on its first detach, then detaches within the same close. Both fail on `main`. The existing APFS, retained-attachment and attach-cleanup cases still pass. - The real-`hdiutil` test "classifies a real hdiutil verify refusal…" fails on this Mac identically on unmodified `main`. It's environmental and passes on CI runners. - Real macOS 26.7 check with this branch: 3 HFS+ and 5 APFS `hdiutil create` images through `rea inspect-artifact`. All succeeded, and `hdiutil info` showed no leftover devices. - `npm run check:fast` No real Hopper or Ghidra verification was run, because this change does not touch those providers. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-17 21:09:05 +00:00