harness/core
5
0
Fork 1
Code Issues Pull requests 15 Projects Releases 238 Packages 1 Wiki Activity Actions

fix: parallel dispatcher workers and direct Main queries #328

Merged
karsten merged 1 commit from fix/dispatcher-concurrency into main 2026-11-11 05:55:49 +00:00
Conversation 2 Commits 1 Files changed 13 +125 -16
karsten commented 2026-11-11 05:55:25 +00:00
Owner
Copy link

Problem

Default worker capacity 1 serializes independent dispatcher tasks; ambiguous delegationPolicy can make Main unnecessarily route short questions through workers.

Changes

  • Default to 4 active workers across Core, starter and setup; preserve explicit operator/backend limits.
  • Distinguish optional delegation from preferred route; direct answers and bounded lookups stay in Main.
  • Supply current worker capacity/active/queued host facts without changing permissions or child restrictions.
  • Document fleet-wide rollout and explicit capacity overrides.

Verification

  • Full suite: 1800 passed, 1 skipped, 0 failed.
  • Final targeted suite: 24 passed; TypeScript and git diff --check clean.
  • Regression coverage: independent review completes while publisher runs; saturated pool accepts queued tasks and refills on settlement; Main retains read/bash tools.

Activation

Requires released Core rollout and controlled process restart. Audit explicit capacity: 1 overrides fleet-wide; no admission bypass or cancellation of active work.

Operator explicitly requested push, PR, merge and fleet rollout.

## Problem Default worker capacity 1 serializes independent dispatcher tasks; ambiguous `delegationPolicy` can make Main unnecessarily route short questions through workers. ## Changes - Default to 4 active workers across Core, starter and setup; preserve explicit operator/backend limits. - Distinguish optional delegation from preferred route; direct answers and bounded lookups stay in Main. - Supply current worker capacity/active/queued host facts without changing permissions or child restrictions. - Document fleet-wide rollout and explicit capacity overrides. ## Verification - Full suite: 1800 passed, 1 skipped, 0 failed. - Final targeted suite: 24 passed; TypeScript and git diff --check clean. - Regression coverage: independent review completes while publisher runs; saturated pool accepts queued tasks and refills on settlement; Main retains read/bash tools. ## Activation Requires released Core rollout and controlled process restart. Audit explicit capacity: 1 overrides fleet-wide; no admission bypass or cancellation of active work. Operator explicitly requested push, PR, merge and fleet rollout.
karsten commented 2026-11-11 07:08:22 +00:00
Author
Owner
Copy link

Rollout checkpoint (operator-authorized fleet rollout): source merged and active in released Core 0.1.229 / 057cc37b88, including subsequent parallel Core fixes.

Live verified: Che, Randy, Minato, Shikamaru, Akira, Kakashi, SaaStaq Moist-Hayate and Odo. Six authenticated capabilities checks (including native Akira PID/path qualification); Moist/Odo intentionally have no API: matching live build, Matrix connected, fresh sync checkpoints verified. Minato and both local Shikamaru profiles now have worker capacity 4; Che 2/Randy 3/Akira 2 retained as existing explicit limits.

Not full convergence: native Karsten agent remains on 0.1.223. Existing owner attempted preparation/activation and safely deferred: GET admission reports unavailable with blocker unsupported update owner: gateway slack; POST is HTTP 503. No admission bypass or forced restart. Peter service/timer disabled; Severus runtime absent at inventoried host; Randy-next/workflow-reviewer deployments absent; none recreated. Kubernetes Karsten remains replicas=0.

Followups merged with green checks: Overlays #324 (Minato capacity), #325 (test-local clock isolation), #331 (empty Forgejo check inventory for explicitly enrolled no-required-check targets). Promotion PRs normally integrated with real maintainer custody and exact source/digest checks. Che had temporary outage during Recreate + transient node DNS image-pull failure; recovered and API verified. Underlying DNS issue is not claimed fixed. Kakashi retains unrelated Argo OutOfSync despite matching live Core/healthy API.

Private detailed receipt: ~/.cache/hayate-maintenance/dispatcher-concurrency/rollout-summary.json. Native Akira receipt: ~/.local/share/hayate-agent/dispatcher-rollout/activation.json on work.

Rollout checkpoint (operator-authorized fleet rollout): source merged and active in released Core 0.1.229 / 057cc37b88e7f3114686233abec147c6eac70985, including subsequent parallel Core fixes. Live verified: Che, Randy, Minato, Shikamaru, Akira, Kakashi, SaaStaq Moist-Hayate and Odo. Six authenticated capabilities checks (including native Akira PID/path qualification); Moist/Odo intentionally have no API: matching live build, Matrix connected, fresh sync checkpoints verified. Minato and both local Shikamaru profiles now have worker capacity 4; Che 2/Randy 3/Akira 2 retained as existing explicit limits. Not full convergence: native Karsten agent remains on 0.1.223. Existing owner attempted preparation/activation and safely deferred: GET admission reports unavailable with blocker `unsupported update owner: gateway slack`; POST is HTTP 503. No admission bypass or forced restart. Peter service/timer disabled; Severus runtime absent at inventoried host; Randy-next/workflow-reviewer deployments absent; none recreated. Kubernetes Karsten remains replicas=0. Followups merged with green checks: Overlays #324 (Minato capacity), #325 (test-local clock isolation), #331 (empty Forgejo check inventory for explicitly enrolled no-required-check targets). Promotion PRs normally integrated with real maintainer custody and exact source/digest checks. Che had temporary outage during Recreate + transient node DNS image-pull failure; recovered and API verified. Underlying DNS issue is not claimed fixed. Kakashi retains unrelated Argo OutOfSync despite matching live Core/healthy API. Private detailed receipt: ~/.cache/hayate-maintenance/dispatcher-concurrency/rollout-summary.json. Native Akira receipt: ~/.local/share/hayate-agent/dispatcher-rollout/activation.json on work.
karsten commented 2026-11-11 07:39:48 +00:00
Author
Owner
Copy link

Previously reported native Karsten Slack blocker is now resolved: Plugins #85 merged and exact immutable module f13f39b installed/active. Core 0.1.229 API healthy; actual POST update-admission returns HTTP 201 admitted with blockers=[], verification lease released. Nine active installations now have the dispatcher/Main fix. Native Karsten has 39 passing native Slack acceptance tests. Additional strict-settings reload defect fixed and merged in Core #334, full suite 1834 passed/1 skip; its subsequent release rollout is separate and not yet claimed live.

Previously reported native Karsten Slack blocker is now resolved: Plugins #85 merged and exact immutable module f13f39b installed/active. Core 0.1.229 API healthy; actual POST update-admission returns HTTP 201 admitted with blockers=[], verification lease released. Nine active installations now have the dispatcher/Main fix. Native Karsten has 39 passing native Slack acceptance tests. Additional strict-settings reload defect fixed and merged in Core #334, full suite 1834 passed/1 skip; its subsequent release rollout is separate and not yet claimed live.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
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
harness/core!328
No description provided.