feat: Recover interrupted workflow runs safely

No labels

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

dries/ocman#321

dries commented

2026-07-13 19:46:13 +02:00

Owner

Parent

#311

What to build

Reconcile active runs after ocman restarts without duplicating uncertain side effects. Reconnect live agent work where possible, replay only explicitly retry-safe attempts, and pause branches that require a human decision.

Acceptance criteria

Startup discovers workflow attempts left in dispatching/running/canceling states and reconciles them before normal scheduling.
Still-live OpenCode sessions are reattached to their attempts and continue reporting state.
Retry-safe interrupted attempts may create a new attempt automatically under existing limits.
Interrupted attempts with uncertain side effects transition to unknown and dependent scheduling pauses.
Users can resolve unknown attempts as succeeded, failed, or eligible for retry, with an audit record.
Resource and workspace leases are retained, recovered, or released based on reconciled executor state.
No restart path dispatches two active attempts for the same node attempt slot.
Pause stops new scheduling while active attempts finish; cancel requests termination and marks unreachable descendants skipped.
Recovery and resolution states are visible and controllable from REST and the UI.
Tests recreate the service/database at each interruption boundary and verify no blind side-effect replay.

Blocked by

## Parent #311 ## What to build Reconcile active runs after ocman restarts without duplicating uncertain side effects. Reconnect live agent work where possible, replay only explicitly retry-safe attempts, and pause branches that require a human decision. ## Acceptance criteria - [ ] Startup discovers workflow attempts left in dispatching/running/canceling states and reconciles them before normal scheduling. - [ ] Still-live OpenCode sessions are reattached to their attempts and continue reporting state. - [ ] Retry-safe interrupted attempts may create a new attempt automatically under existing limits. - [ ] Interrupted attempts with uncertain side effects transition to unknown and dependent scheduling pauses. - [ ] Users can resolve unknown attempts as succeeded, failed, or eligible for retry, with an audit record. - [ ] Resource and workspace leases are retained, recovered, or released based on reconciled executor state. - [ ] No restart path dispatches two active attempts for the same node attempt slot. - [ ] Pause stops new scheduling while active attempts finish; cancel requests termination and marks unreachable descendants skipped. - [ ] Recovery and resolution states are visible and controllable from REST and the UI. - [ ] Tests recreate the service/database at each interruption boundary and verify no blind side-effect replay. ## Blocked by - #314 - #315 - #319 - #320

dries added the

label

dries referenced this issue

2026-07-13 19:46:15 +02:00

feat: Ship the adversarial migration workflow #326

dries referenced this issue from a commit

2026-07-16 00:48:06 +02:00

feat: recover interrupted workflow runs safely (#321)

dries closed this issue

2026-07-16 00:56:09 +02:00