Measure And Right-Size An Euler Job

Use the compact field guide above while reading your existing job. Requested resources describe its reservation, not measured use. The tiny training job locates the fields; the separate fictional serial run supplies the sizing exercise.

Look forYou can distinguish allocated resources from measured use.

From the local terminal, connect with the tested euler alias. Run every later command in the Euler shell that opens.

ssh euler

ssh euler

ssh euler

Look forThe prompt changes to an Euler login node.

If this differsReturn to the SSH access mission; do not run Euler commands in the local shell.

Read the job ID saved by the previous mission. If the file is missing, list recent accounting records and identify the single passport-cpu job instead of submitting another job.

id_file="$HOME/passport-euler/first-job.id"
if [ -s "$id_file" ]; then printf 'Stored job ID: '; cat "$id_file"; else sacct -X -u "$USER" --starttime today --format=JobID,JobName,State,Elapsed; fi

Look forYou identify the existing passport-cpu job ID.

If this differsStop if you cannot identify one unambiguous training job.

Load the stored ID. If it was missing, enter the passport-cpu ID you identified and save it. Then query state, exit code, elapsed time, allocated CPUs, requested memory, and maximum resident memory. Continue only after the main job is COMPLETED with exit code 0:0.

(
set -eu
id_file="$HOME/passport-euler/first-job.id"
if [ -e "$id_file" ] && [ ! -s "$id_file" ]; then printf 'STOP: %s exists but is empty. Ask for help before changing it.\n' "$id_file" >&2; exit 1; fi
if [ -s "$id_file" ]; then job_id="$(cat "$id_file")"; else read -r -p 'Existing passport-cpu job ID: ' job_id; fi
case "$job_id" in ''|*[^0-9]*) printf 'STOP: job ID must contain digits only.\n' >&2; exit 1;; esac
if [ ! -e "$id_file" ]; then mkdir -p "$(dirname "$id_file")"; printf '%s\n' "$job_id" > "$id_file"; chmod 600 "$id_file"; fi
sacct -j "$job_id" --format=JobID,JobName,State,ExitCode,Elapsed,AllocCPUS,ReqMem,MaxRSS
state="$(sacct -X -n -j "$job_id" --format=State | awk 'NF {print $1; exit}' | cut -d+ -f1)"
exit_code="$(sacct -X -n -j "$job_id" --format=ExitCode | awk 'NF {print $1; exit}')"
[ "$state" = COMPLETED ] && [ "$exit_code" = 0:0 ] || { printf 'STOP: wait for COMPLETED 0:0 or inspect the failed job; do not resubmit. Current result: %s %s\n' "${state:-unavailable}" "${exit_code:-unavailable}" >&2; exit 1; }
printf 'accounting-ready\n'
)

Look forsacct shows the main job and its steps, the main job is COMPLETED with exit code 0:0, MaxRSS is read from the relevant step when available, and the final line is accounting-ready.

If this differsWait for accounting to appear or verify the job ID; do not resubmit.

Use seff for a concise CPU and memory efficiency summary, while keeping sacct as the source for exact fields.

(
job_id="$(cat "$HOME/passport-euler/first-job.id")"
case "$job_id" in ''|*[^0-9]*) printf 'STOP: stored job ID is invalid.\n' >&2; exit 1;; esac
seff "$job_id"
)

Look forseff reports the same job's CPU and memory efficiency.

If this differsUse sacct and application logs if seff is unavailable.

Compare allocated CPU count, requested memory and time with measured use. The tiny job checks the workflow and field lookup; its utilization is not representative of a production input.

Look forYou can explain AllocCPUS, ReqMem, MaxRSS, Elapsed, State, and ExitCode.

Use the representative serial-run scenario in the questions, not the tiny calculation, to choose the next test. Before reading the model, compare reserved and used resources and decide what headroom is justified. This lesson asks for a decision; it does not submit that request.

Look forYour selected next test has a reason for its CPU, total memory and time values.

Enter the existing job ID and confirm that you personally inspected sacct and seff. Do not paste logs into the public record.

Look forThe result contains only the requested inspection facts and no username, path, log, or other private value.