allaun
e5bfdc0492
Containerized LKB worker (Dockerfile.lkb), Postgres-backed queue, auto-deploy timer, and eigensolid convergence check module. Build: 0 jobs (no Lean build needed)
60 lines
No EOL
1.7 KiB
Bash
Executable file
60 lines
No EOL
1.7 KiB
Bash
Executable file
#!/usr/bin/env bash
|
|
set -euo pipefail
|
|
|
|
HOST=$(uname -n)
|
|
case "$HOST" in
|
|
*neon*) DEFAULT_WORKERS=12 ;;
|
|
*) DEFAULT_WORKERS=4 ;;
|
|
esac
|
|
|
|
WORKERS="$DEFAULT_WORKERS"
|
|
ROLE="worker"
|
|
BUILD=false
|
|
PLATFORM=""
|
|
|
|
while [[ $# -gt 0 ]]; do
|
|
case "$1" in
|
|
--workers) WORKERS="$2"; shift 2 ;;
|
|
--role) ROLE="$2"; shift 2 ;;
|
|
--build) BUILD=true; shift ;;
|
|
--platform) PLATFORM="$2 --platform $2"; shift 2 ;;
|
|
*) echo "Usage: $0 [--workers N] [--role seed|worker|all] [--build] [--platform linux/arm64|linux/amd64]"
|
|
echo " Default workers: $DEFAULT_WORKERS"; exit 1 ;;
|
|
esac
|
|
done
|
|
|
|
DIR="$(cd "$(dirname "$0")" && pwd)"
|
|
IMAGE="lkb-worker:latest"
|
|
|
|
if [ "$BUILD" = true ] || ! podman image exists "$IMAGE" 2>/dev/null; then
|
|
echo "[build] $IMAGE ..."
|
|
podman build -t "$IMAGE" -f "$DIR/Dockerfile.lkb" "$DIR"
|
|
echo "[build] done"
|
|
fi
|
|
|
|
run_seed() {
|
|
podman rm -f lkb-seed 2>/dev/null || true
|
|
podman run -d --rm --name lkb-seed --network host "$IMAGE" seed
|
|
echo "[seed] started. Logs: podman logs -f lkb-seed"
|
|
}
|
|
|
|
run_workers() {
|
|
echo "[worker] launching $WORKERS containers ..."
|
|
for i in $(seq 1 "$WORKERS"); do
|
|
podman rm -f "lkb-worker-$i" 2>/dev/null || true
|
|
podman run -d --rm --name "lkb-worker-$i" --network host "$IMAGE" worker
|
|
done
|
|
echo "[worker] $WORKERS launched"
|
|
}
|
|
|
|
case "$ROLE" in
|
|
seed) run_seed ;;
|
|
worker) run_workers ;;
|
|
all) run_seed; sleep 3; run_workers ;;
|
|
esac
|
|
|
|
echo "=== Running ==="
|
|
podman ps --format "table {{.Names}} {{.Status}}" | grep lkb-
|
|
echo "=== DB queue ==="
|
|
PGPASSWORD=kaleidoscope psql -h 100.79.14.103 -U kaleidoscope -d research_stack \
|
|
-c "SELECT status, count(*) FROM lkb.compute_queue GROUP BY status ORDER BY status" 2>/dev/null || true |