Repository navigation
🐛 fix: serve continuous re-kicks between busy cycles - #14824
Conversation
Check due continuous re-kicks between loop iterations and run a cached eval without maintenance sweeps. Defer remaining same-agent kicks after delivery or a live-turn bounce, without recording busy deferrals as governor kick failures. Co-authored-by: clubanderson <407614+clubanderson@users.noreply.github.com> Hive-Run: hivecommons#14820 Hive-Plan: continuous-rekick-wake-and-busy-deferral Hive-Spec: issue-14820#expected Signed-off-by: thatguywob <thatguywob@users.noreply.github.com>
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
There was a problem hiding this comment.
Read the diff against the v5 tree and the callers; the change does what #14820's "Expected" section asks and I found no defect. Looks correct to me.
What I verified:
src/cmd/hive/main.gorun loop:serveContinuousWake()now runs at the top of every iteration before theselect, so a due re-kick is served whichever channel woke the loop. Hot-loop risk is bounded by the existingcontinuousWakeMinGapcheck incontinuousWakeDecision(src/cmd/hive/continuous_wake.go:72) andlastWake = now; the stall backoff viawakeHold.observeis retained. The lean cycle takescfgReloadMuliketickdoes and intentionally skipsrestartCrashed, watchdog, rotation and sweeps — the cadence tick still runs those.src/crux/agent/manager_kick.go:196,290wrap the not-ready error incrux.ErrAgentBusyonly whenkickUndeliverableTranscriptLiveness(busy_visibility.go:235) says the transcript is live;m.muis held at both call sites askickNotReadyErrorLockedrequires.kickDispatchDeps.sendKickis wired directly toagentMgr.SendKick(main.go:7678), soerrors.Isineval_cycle_seams.go:582,596sees the sentinel unwrapped.eval_cycle_seams.go:596: a busy bounce ends the span withniland skipsonFailed, soRecordKickFailureno longer doublescontinuousBackofffor a merely busy agent.
One thing worth a maintainer's glance (not a defect): this deliberately replaces the contract pinned by the old TestDispatchAgentKicks_SameAgentStaysOrdered — after a successful fix kick, the governor kick to the same agent in that cycle is now deferred rather than typed in behind it. Given the old behaviour was a guaranteed 120s prompt-timeout failure per the issue, the deferral is the right call, and the test rename makes the change explicit.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
|
Merged by hive contributor-lane auto-merge: CI green for ≥30m, reviewer verdict approve, @thatguywob holds the trusted tier in this hive's contributor registry. |
Summary
Related issues
Fixes #14820
Testing
gofmt -lon all changed Go files: no output.Contributor checklist
changelog.d/fixed-14820-continuous-rekick.md.v5; title follows the repository emoji convention.— hive: backend=codex
🐝 Hive Agent:
contributor| SHA:256ef585f