[hanhainebula]

Summary

On busy hosts, Docker/containerd can take more than 60 seconds to create or remove containers. The shared runner can therefore fail a gateway preflight or teardown even when the operation would complete successfully with a longer bounded wait.

Introduce a shared 300-second lifecycle timeout for gateway preflight container execution, preflight container removal, and docker compose down. Task execution budgets and egress leases are unchanged. Existing ownership checks and cleanup failure handling remain in place.

Update the cleanup timeout assertions and add a regression that simulates slow teardown, checking that the private runtime directory remains available until teardown succeeds and is then removed.

This is the first of two related PRs. The follow-up PR revises three retrieval task packages; this PR contains only the shared lifecycle fix and its tests.

Commit: 1bd88b57237c0a2276c8e915d3f6cf6defdd8d5c — Allow bounded slow Docker lifecycle operations.

Changed files:

• scripts/harbor_environments.py
• scripts/tests/test_egress_adapter.py

Validation evidence

Checks ran in an isolated worktree based on fork main at 21ab7a32ed75d4d463b6bc91deb0f304bcb3e3b5, containing only these two file changes:

python -m unittest discover -s scripts/tests -p 'test_*.py'
python scripts/check_release.py
git diff --check

• All 175 shared tests passed, including the new slow-teardown regression.
• Release static checks reported zero errors and zero warnings.
• The diff check passed, and the commit scope was verified to contain exactly the two files above.

The slow-teardown case uses mocks; these checks do not demonstrate a real Docker operation taking more than 60 seconds. Live Docker/model trials were not rerun for this commit. Integration and CI checks against the current upstream base remain part of PR review.

Contributor checklist

• Changes are limited to shared Docker lifecycle waits and their regression tests.
• Task budgets, scoring, network leases, and task packages are unchanged.
• No secrets, downloaded assets, or job outputs included.
• Validation results and their limits recorded.
• Enable Allow edits from maintainers, or agree on contributor-applied review fixes.

Maintainer merge checklist

• Review the lifecycle timeout and cleanup behavior.
• Run required checks against the current upstream base and final PR head.
• Merge before the follow-up retrieval-task calibration PR, preserving contributor history.