Repository navigation
runner: a firing that fails gives back the tokens it took - #38
Merged
Merged
Conversation
Every actor kind took its inputs before it ran -- an action after its guard, a tool and an agent on entry -- so a firing that raised had lost them, and no checkpoint saved after the failure could bring them back. Now every dequeue made during a firing is logged per thread, and when the firing raises each token goes back to the head of its queue, in order, before the error propagates: the run stops where the failed firing had not happened. This is phase 1 of docs/proposals/resume.md, and the log is what a later phase records as a step's `consumed`. A nested workflow and an if / loop are not wrapped. One firing of theirs runs other actors' firings to completion, each atomic on its own, so giving back theirs too would hand a token to a firing that had already consumed it. A firing that succeeded keeps what it took, whatever fails after it. Claude-Session: https://claude.ai/code/session_015VK7fH1c4aKbexnq2QcuKU
This was referenced Oct 1, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Phase 1 of
docs/proposals/resume.md(#37).Every actor kind took its inputs before it ran: an action after its guard, a tool and an agent on entry. A firing that raised had therefore lost them, and no checkpoint saved after the failure could bring them back.
How: every
Queue.dequeue()/try_dequeue()made during a firing is logged per thread (_atomic_firing, around_step_actor). When the firing raises, each token goes back to the head of its queue in reverse order, so the queue is as it was. This is safe without holding the queues: one actor never fires twice at once, and each queue has one consumer. A producer appending to the tail is unaffected. The log is what phase 3 will record as a step'sconsumed.Not wrapped: a nested workflow and an
if/loop. One firing of theirs runs other actors' firings to completion, each atomic on its own. Giving back theirs too would hand a token to a firing that had already consumed it. A firing that succeeded keeps what it took, whatever fails after it.Not restored: the actor's own state (
self._done = Truebefore a raise). That is the checkpoint's job, in phase 2.Tests:
tests/test_atomic_firings.py(8). It covers an action (single and multi-token), a tool, an agent, a failure inside a nested workflow (the token is held once, by the child), a failure inside a loop body after an earlier body firing succeeded, and the queue-level semantics. Without the give-back, 6 of them fail. Full suite: 634 passed, 9 skipped.https://claude.ai/code/session_015VK7fH1c4aKbexnq2QcuKU