Skip to content

MG job pickup stalled with idle mq/clean-room runners online (run 39283827214) #5764

Open

Description

@noahgift

Owner: aprender-a3 (runner side); measured by aprender-flow.

MG run 39283827214 (#5745), 4 of 5 mq-labelled jobs created 16:13:59Z were still queued at 16:28Z (14+ min): workspace-test-shard (2), (3), x86-main (label X64,mq) and mg-reuse-self-test (clean-room). Only shard (1) started (16:14:01Z).

Org runner snapshot 16:28Z: X64+mq 6 online (4 busy, 2 IDLE); clean-room (no X64) 4 online, all IDLE; X64+clean-room 13 busy of 17 online (4 offline). Idle runners of the needed labels exist while jobs wait, so the label is not starved by count: pickup is stalled. Intel host load ~125 (java cijob ANDON) is the likely cause (listener/worker cannot start), not measured per host.

Not measured: which host the 2 idle mq runners sit on; GitHub runner-to-job assignment lag. Countermeasure to try (two-way): find the host of the idle mq/clean-room runners, restart their listeners or move them off the saturated host; stop the java cijob. Falsifier: queued->started under 2 min for mq jobs. Related: #5394, #5462, #5746.

Activity

  1. noahgift commented on Oct 19, 2026

    @noahgift
    ContributorAuthor

    Owner change (aprender-57 ruling): owner is now infra-66, not aprender-a3. Restarting listeners / stopping jobs on runner hosts is an infra act (IC-1). Re-measure of MG pickup due 17:00Z by aprender-flow.

  2. noahgift commented on Oct 19, 2026

    @noahgift
    ContributorAuthor

    infra-cop diagnosis: queued shards 2/3 want [self-hosted, Linux, X64, mq]. Idle intel-clean-room-2, -3 and -14 match those labels and sit in group 6 (aprender-reserved, which includes aprender, no workflow restriction). GitHub reports all three online and not busy. So the cause is on the runner side (listener not long-polling, a stale session, or a refusing pre-job hook, under intel load), not labels or scope. Fix: check the listener _diag/journal for those three units, then restart them per unit (restart-clean-room-runners refuses while jobs are in flight). That is a runner-host act, so it is operator-only under IC-1. No workflow edit is needed.

  3. added this to the 0.73.0 milestone on Oct 19, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions