Each daemon restart re-queues running tasks from scratch (one task was re-queued 33 times), paying for the same model work again. Done when a re-queued task resumes from its last completed step, and re-queues after restarts are capped before the task parks with a clear reason.
Each daemon restart re-queues running tasks from scratch (one task was re-queued 33 times), paying for the same model work again. Done when a re-queued task resumes from its last completed step, and re-queues after restarts are capped before the task parks with a clear reason.
Fixed in e75d5c8ad. Crash recovery now re-queues a running task with resume_from set to the interrupted root step, the results of the last step it finished, and keep_worktree. A crash inside the first resumed step keeps its resume point, because the run clears it only after that step finishes. A task re-queued by 3 restarts in a row parks as needs_decision with a reason; an operator retry or requeue resets the count (new column restart_requeues, migration 0071). The decision is a pure function with one table-driven test. task check was clean. If a crash lands between two steps with none running, the task still starts from the first step.
Fixed in e75d5c8ad. Crash recovery now re-queues a running task with resume_from set to the interrupted root step, the results of the last step it finished, and keep_worktree. A crash inside the first resumed step keeps its resume point, because the run clears it only after that step finishes. A task re-queued by 3 restarts in a row parks as needs_decision with a reason; an operator retry or requeue resets the count (new column restart_requeues, migration 0071). The decision is a pure function with one table-driven test. task check was clean. If a crash lands between two steps with none running, the task still starts from the first step.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Each daemon restart re-queues running tasks from scratch (one task was re-queued 33 times), paying for the same model work again. Done when a re-queued task resumes from its last completed step, and re-queues after restarts are capped before the task parks with a clear reason.
Fixed in
e75d5c8ad. Crash recovery now re-queues a running task with resume_from set to the interrupted root step, the results of the last step it finished, and keep_worktree. A crash inside the first resumed step keeps its resume point, because the run clears it only after that step finishes. A task re-queued by 3 restarts in a row parks as needs_decision with a reason; an operator retry or requeue resets the count (new column restart_requeues, migration 0071). The decision is a pure function with one table-driven test. task check was clean. If a crash lands between two steps with none running, the task still starts from the first step.