fix: release conversation subject in cancel terminal update

[onatozmenn]

HUMAN:

I ran the targeted pytest suite and confirmed all 109 tests pass, including the two new subject-release cases that fail on base and pass with the fix, with ruff checks also clean.

---

Why

Cancelling a run that holds a conversation subject only released it when the run had recorded a sandbox_id (and in a second write after the terminal update). Local-mode and conversation-turn child runs often have no sandbox_id, so subject_released_at stayed unset and a resubmitted subject kept deduplicating against the cancelled run instead of creating a replacement (#472).

Summary

• Record subject_released_at atomically in the cancel terminal UPDATE (conditional on subject_key), independent of sandbox_id.
• Drop the now-redundant post-update release; sandbox cleanup path unchanged.
• Add regression tests: subject run without sandbox releases, ordinary runs untouched, cancelled subject routes resubmission to a new run via continue_conversation.

Issue Number

Fixes #472

How to Test

python -m pytest tests/test_cancel_run.py tests/test_conversations.py tests/test_watchdog.py -q — 109 passed. The two new subject tests fail on base and pass with the fix. ruff check / ruff format --check clean.

Type

• Bug fix
• Feature
• Refactor
• Breaking change
• Docs / chore

[all-hands-bot]

👋 This PR needs a couple of things fixed before OpenHands can review it:

• the PR description's HUMAN: section needs at least 20 characters describing what you tested, not just the template placeholder

Push an update once this is addressed and this check re-runs automatically.

This is an automated check - no AI was used to generate this comment.

[all-hands-bot]

👋 This PR needs a couple of things fixed before OpenHands can review it:

• the PR description's HUMAN: section needs at least 20 characters describing what you tested, not just the template placeholder

Push an update once this is addressed and this check re-runs automatically.

This is an automated check - no AI was used to generate this comment.

[Copilot]

🟡 Changes recommended

The regression test does not exercise an actual resubmission after cancellation.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

This pull request fixes cancellation so conversation subjects are released atomically, including runs without sandboxes.

Changes:

• Updates cancellation to set subject_released_at.
• Removes redundant post-update release logic.
• Adds cancellation regression tests.

File summaries

File	Description
tests/test_cancel_run.py	Tests subject release and ordinary cancellations.
openhands/automation/router.py	Atomically releases subjects during cancellation.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[onatozmenn]

Good catch — the resubmission test now goes through the real routing path: it cancels a queued subject run, then calls continue_conversation for the same subject and asserts needs_run is True. On base this fails (the turn gets coalesced into the dead run, so no replacement run is ever created); with the fix it passes.