Files
jspace/scripts/gpuq_worker.sh
T
2026-07-14 13:00:38 +02:00

42 lines
1.5 KiB
Bash
Executable File

#!/bin/bash
# Per-GPU work queue worker, fed from the S3 bucket (rclone remote "jspace").
#
# Usage: gpuq_worker.sh <worker-id> <gpu-index>
# e.g.: gpuq_worker.sh node1-gpu3 3
# Run ONE per GPU, each in its own tmux window:
# tmux new-window -t ssh_tmux -n gpuq3 'bash ~/jspace/scripts/gpuq_worker.sh node1-gpu3 3'
#
# Queue layout (per worker, race-free by construction — no work stealing):
# jspace:jspace/gpuq/<worker-id>/pending/<job>.sh submitted jobs
# jspace:jspace/gpuq/<worker-id>/done/<job>.sh|.log finished jobs + logs
# Jobs run oldest-first by name — prefix with 00_, 01_, ... to order them.
#
# Job contract: plain bash, repo at ~/jspace, venv at ~/jspace/.venv (or
# /venv/main on vast images — job script picks), CUDA_VISIBLE_DEVICES is set
# by the worker. Job handles its own LOOP_OUT / result rclone if needed.
set -u
ID=$1
GPU=$2
Q="jspace:jspace/gpuq/$ID"
LOCAL="$HOME/gpuq/$ID"
mkdir -p "$LOCAL/pending"
echo "[gpuq $ID] worker up on GPU $GPU, polling $Q/pending"
while true; do
rclone move "$Q/pending" "$LOCAL/pending" --include "*.sh" -q 2>/dev/null
job=$(ls "$LOCAL"/pending/*.sh 2>/dev/null | sort | head -1)
if [ -z "$job" ]; then
sleep 20
continue
fi
name=$(basename "$job" .sh)
echo "[gpuq $ID] $(date +%F_%H:%M:%S) START $name"
CUDA_VISIBLE_DEVICES=$GPU bash "$job" 2>&1 | tee "$LOCAL/$name.log"
rc=${PIPESTATUS[0]}
echo "[gpuq $ID] $(date +%F_%H:%M:%S) DONE $name rc=$rc"
echo "rc=$rc" >> "$LOCAL/$name.log"
rclone copy "$LOCAL/$name.log" "$Q/done/" -q
rclone moveto "$job" "$Q/done/$name.sh" -q
done