Skip to content

Harden Docker Sandboxes recovery and support v0.42.1 - #44

Merged
joecklau merged 2 commits into
developfrom
fix/docker-sandboxes-recovery-hardening
Sep 8, 2026
Merged

Harden Docker Sandboxes recovery and support v0.42.1#44
joecklau merged 2 commits into
developfrom
fix/docker-sandboxes-recovery-hardening

Conversation

@joecklau

@joecklau joecklau commented Sep 8, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Harden Docker Sandboxes control-plane recovery and lifecycle reconciliation with durable host-wide coordination, exact identity checks, and capacity-preserving retries.
  • Reconcile historical sandbox identities only after independent exact-absence verification, and expose redacted provisioning failure causes.
  • Support the inert default SSH-agent environment injected by sbx v0.42.1 while retaining rejection of actual agent endpoints and unexpected forwarding configuration.
  • Export complete local-build SBOM evidence through ordinary files to avoid BuildKit attestation size limits, with bounded validation and artifact association.
  • Add regression coverage and update operational and security documentation.

Validation

  • Relevant Go test packages passed: internal/image, internal/pool, internal/provider/dockersandboxes, internal/prebuilt, and cmd/epar-prebuilt-publisher. Tests requiring local listeners were run outside the sandbox with a canonical temporary directory.
  • Local Full arm64 template built and passed live admission on sbx v0.42.1.
  • Live two-runner lifecycle verification exercised replacement and cleanup.
  • Operator reports three successful CI jobs with runner recycling and replacement using the fixed local image.

Remaining verification and follow-ups

  • Published prebuilt-image verification remains pending; the normal configuration uses full-latest, whose publication requires a Full schedule or explicit Full workflow dispatch.
  • A startup host-trust refresh cancellation warning remains a separate follow-up; this PR does not claim to resolve it.
  • Separate September 2–8 resource-exhaustion concerns are not claimed resolved by this change.
  • Local configuration, credentials, and investigation evidence are excluded.

Includes commits 67e2b8a and 5ff0ec7. Targets develop; this PR does not merge or publish the stable image.

Coordinate bounded daemon recovery through a durable host-wide lease and ledger. Preserve uncertain instance ownership and capacity, verify exact inventory and cleanup, and separate architecture admission deadlines.

Add recovery regression coverage and recognize exact sbx v0.42 missing-sandbox diagnostics while preserving legacy handling. Document operational boundaries and pending v0.42.1 live validation.
@joecklau
joecklau deployed to epar-live-ci September 8, 2026 17:03 — with GitHub Actions Active
@joecklau
joecklau merged commit 9550b06 into develop Sep 8, 2026
29 of 30 checks passed
@joecklau
joecklau deleted the fix/docker-sandboxes-recovery-hardening branch September 8, 2026 18:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants