Skip to content

ci: run Frontier CI on the g1 partition, and raise the queue-wait budget to 4h #1767

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
sbryngelson merged 3 commits into from
Aug 27, 2026
+17 −4
Merged

ci: run Frontier CI on the g1 partition, and raise the queue-wait budget to 4h #1767

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
8f66680
ci(frontier): submit CI jobs to the g1 partition
sbryngelson Aug 26, 2026
8219724
Merge 'ci/phoenix-queue-timeout-combined-alloc' (#1763) into the g1 b…
sbryngelson Aug 27, 2026
15177d9
ci: raise the queue-wait budget to 4h for all clusters
sbryngelson Aug 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 9 additions & 1 deletion .github/scripts/monitor_slurm_job.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,7 +86,15 @@ is_terminal_state() {
# (Phoenix 'embers') a job can stay PENDING for hours, burning the CI job
# timeout and holding a runner slot; fail early so it reads as queue starvation,
# not a test failure. 0 = wait indefinitely.
: "${SLURM_MAX_QUEUE_SECONDS:=5400}" # 90 minutes
#
# The budget has to clear a normal bad day on a busy machine, or it converts
# routine queue pressure into red CI. Frontier's own numbers make the case:
# over one week, MFC jobs on `batch` waited p50=1m but p90=96m and p95=176m,
# with a 466m tail. A 90-minute budget cut into that distribution, tripping on
# 11% of the jobs that did eventually start -- plus the ones that never did.
# Four hours clears p95 with room to spare while staying well inside the 480m
# job-level `timeout-minutes`, which remains the real backstop.
: "${SLURM_MAX_QUEUE_SECONDS:=14400}" # 4 hours
# Reject a non-integer override rather than silently skipping the budget.
if ! [[ "$SLURM_MAX_QUEUE_SECONDS" =~ ^[0-9]+$ ]]; then
echo "ERROR: SLURM_MAX_QUEUE_SECONDS must be a non-negative integer (seconds), got '$SLURM_MAX_QUEUE_SECONDS'" >&2
Expand Down
11 changes: 8 additions & 3 deletions .github/scripts/submit-slurm-job.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,7 +62,10 @@ case "$cluster" in
# CFD154; submitting under it now fails outright with "Invalid qos
# specification". "normal" is the only QOS on this allocation without a
# one-job-at-a-time cap, so it is the only one that can run the CI
# matrix concurrently.
# matrix concurrently. Note that the g1 partition carries its own
# partition QOS (also named "g1"), which slurmctld applies on its own
# when a job lands there. Do not add --qos=g1: CFD154 has no
# association with it and sbatch rejects the job outright.
qos="normal"
# Let each job's slurmstepd broker its own steps instead of routing
# every srun through slurmctld. The in-job test suite launches ~1700+
Expand Down Expand Up @@ -105,9 +108,11 @@ if [ "$device" = "cpu" ]; then
#SBATCH --mem-per-cpu=8G"
;;
frontier|frontier_amd)
# g1 is a dedicated 64-node carve-out; its nodes are not in batch,
# so CI starts promptly instead of queueing behind the machine.
sbatch_device_opts="\
#SBATCH -n 32
#SBATCH -p batch"
#SBATCH -p g1"
;;
esac
elif [ "$device" = "gpu" ]; then
Expand Down Expand Up @@ -135,7 +140,7 @@ elif [ "$device" = "gpu" ]; then
frontier|frontier_amd)
sbatch_device_opts="\
#SBATCH -n 8
#SBATCH -p batch"
#SBATCH -p g1"
;;
esac
else
Expand Down
Loading