Skip to content

Fleet dispatch hardening: retire raw-nohup hunts (timeout, heartbeat, durable logs, sweep-adoptable records) #426

Description

@aarontrowbridge

Important

Problem — Fleet research dispatch is fire-and-forget: QEC hunts launch as raw ssh <host> 'cd … && nohup … > /tmp/<hunt>.log 2>&1 &' — no timeout, no heartbeat, logs in /tmp (lost on reboot), status discovered by ps grep. A wedged hunt hangs invisibly forever; a reboot orphans it with no trace. The fleet registry's transition table already models crashed/orphaned records — nothing creates records for hunts.

Acceptance Criteria

  • A hunt wrapper: timeout -k bounded, heartbeat file (mtime liveness), logs under ~/.amico/ops/hunts/<id>/ (durable, synced-adjacent)
  • Hunts create fleet records at launch so fleet sweep adopts orphans (pid-liveness guarded — the registry's designed path)
  • Status table reads records, not ps-grep
  • The qec-autoresearch skill's dispatch section rewritten to the wrapper

Design-of-record: vault spec spec-20260818-gpu-plane (ladder item 3). erlich = the-feynmachine (24c + GT 1030 display card — CPU decoder replica; not compute-GPU class).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions