Ilker ManapandClaude Opus 5.5 b8274d6278 A restart is not refused for room another restart is giving back ...

On a 4-core single-machine install, `ctl model resources` was run on the
judge (to two cores) and on the starter (to none), then both restarted.
The judge's restart was refused: its replacement fit on no node even with
its old replica stopped, because the starter still held its two cores. A
restart now waits for other restarts under way (queued, retiring or
draining) when their going would make room, up to DrainGrace.

A drained replica was also kept until its model had enough live replicas,
and a replica being restarted that still answers did not count. So with
replicas restarted one after another, each drained one waited out the
five-minute grace, holding its room.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

2026-10-16 21:33:20 +02:00

..

adopt.go

An adopted on-demand replica counts as used when adopted

2026-10-11 08:55:33 +02:00

demand.go

ComfyUI models of one environment share one replica

2026-10-14 13:55:34 +02:00

evict.go

ComfyUI models of one environment share one replica

2026-10-14 13:55:34 +02:00

facility_test.go

Facility-aware placement: racks, PDUs, holds and power caps

2026-10-05 01:17:49 +02:00

fetch_test.go

Carry a download's progress to where somebody is looking at it

2026-09-27 07:49:08 +02:00

hardware.go

A profile may name the card it needs rather than its class, and fit any site

2026-10-11 16:02:32 +02:00

headroom_test.go

Reserve CPU cores and require memory for CPU models

2026-09-30 17:43:13 +02:00

headroom.go

Facility-aware placement: racks, PDUs, holds and power caps

2026-10-05 01:17:49 +02:00

health.go

ComfyUI models of one environment share one replica

2026-10-14 13:55:34 +02:00

leadership_test.go

A standby that takes over waits for the fleet before placing

2026-10-07 13:40:28 +02:00

measured_test.go

Scale and place by what a replica was measured to serve

2026-09-30 18:21:08 +02:00

measured.go

A profile may name the card it needs rather than its class, and fit any site

2026-10-11 16:02:32 +02:00

notify_test.go

Notifications: webhooks for a parked model, a lost node, a spent budget and new weights; Alertmanager receiver templates

2026-10-04 16:58:11 +02:00

preempt_test.go

Name only the replicas that would make room in headroom

2026-09-30 23:01:40 +02:00

preempt.go

How long an on-demand model holds a card is a live setting

2026-10-14 08:58:38 +02:00

quotas_test.go

Guarantee and limit capacity per part of the tenancy tree

2026-09-30 18:15:38 +02:00

quotas.go

Guarantee and limit capacity per part of the tenancy tree

2026-09-30 18:15:38 +02:00

reconciler_test.go

A restart is not refused for room another restart is giving back

2026-10-16 21:33:20 +02:00

reconciler.go

A restart is not refused for room another restart is giving back

2026-10-16 21:33:20 +02:00

replica.go

A replica says what its engine computes on, and llama.cpp's word on the CPU refuses a GPU placement

2026-10-07 13:02:17 +02:00

restart.go

A restart is not refused for room another restart is giving back

2026-10-16 21:33:20 +02:00

schedule_test.go

Scaling by the clock: schedule windows with their own min and max, prewarm, control.timezone

2026-10-04 23:16:56 +02:00

shared_test.go

ComfyUI models of one environment share one replica

2026-10-14 13:55:34 +02:00

spread_test.go

Pin down where CPU replicas land, and where they deliberately do not

2026-09-16 09:57:12 +02:00

stuck_test.go

Stop an idle model to make room for one somebody asked for

2026-09-17 17:51:07 +02:00

trace_test.go

Tracing beyond the router: reconcile passes, placement and a replica's start traced down to the node

2026-10-05 00:03:17 +02:00

trace.go

Tracing beyond the router: reconcile passes, placement and a replica's start traced down to the node

2026-10-05 00:03:17 +02:00