Runner Health - SGLang CI

Queue Right Now how deep, how long, and how much of it is capacity

Runner Pool	Queued	Split	Wants a runner	Waiting	Runners busy	GPUs	Verdict

Runner Capacity Split runners per runs-on scale set · warm floor to ceiling

Runner Pool	GPU/runner	Min	Max	Split	GPUs	Peak 24h	Now	Check

Shared Model Cache free space on /sgl-data, the HuggingFace cache the whole AMD fleet shares

Measured	Runner Pool	Image	Free	Used	Checkpoint	Job

Runner Pool Status per runs-on label · live pickup

online picking up degraded backlog offline queued, no runner idle no work

Runner Pool	Status	Queued	Max wait	Running	Recent	Runner lost	Runners now	Total avail.

Runner Fleet Report capacity & queue over the last 24h · from AMD CI Job Monitor

Full report

Runner Pool	Peak	Avg	Jobs	Queue p50	p90	p99	Avg run

JOBS STARTED BY HOUR (UTC)

peak busy moderate quiet

00:0023:00

Queue History how long jobs waited for a runner, and how deep the queue got

days and hours are UTC

WAIT FOR A RUNNER (per day · median / p90)

DEMAND & SLO (jobs that queued · picked up < 15m)

QUEUE DEPTH VS RUNNERS BUSY (hourly average · jobs waiting / jobs running)

Runner Pool	Jobs	Picked up < 15m	p50	p90	p99	Queue depth	Running	vs prev

DI CI MI355 Nodes linux-sglang-mi35x-di · Slurm partition amd-sglang

Node	State	GPUs	Host RAM	CI	Held by

Queued Runs 0

view queue on GitHub click to toggle

Run	Workflow	Author	PR / Branch	Runner	Title	Waiting	Link

AMD CI Job Monitor -

View Run

In-Progress Runs 0

Run	Workflow	Author	PR / Branch	Runner	Title	Link

Recent Completed Runs 0

Run	Workflow	Result	Time

Last Passing Runs 0

Run	Workflow	Time