box/bin/swarm-worker-supervise.sh

#!/usr/bin/env bash
#
# swarm-worker-supervise.sh — tmux supervisor for the swarm worker daemon.
#
# Manages a detached tmux session named "swarm-worker" that runs the worker
# main loop and logs to logs/swarm-worker.log.
#
# The worker itself is a PLACEHOLDER until bin/swarm_worker/daemon.py lands.
# When it does, update WORKER_CMD and WORKER_MATCH below.
#
# Usage: swarm-worker-supervise.sh {start|stop|restart|status}
#
set -euo pipefail

SESSION="swarm-worker"
SOCKET_PATH="/tmp/tmux-muse.sock"
LOG_DIR="/home/super/Projects/NetVM/logs"
LOG_FILE="${LOG_DIR}/swarm-worker.log"

# Placeholder worker. Replace with:
#   WORKER_CMD="python3 /home/super/Projects/NetVM/bin/swarm_worker/daemon.py"
# when the real daemon exists. WORKER_MATCH must uniquely match the worker
# process command line; the [a] bracket trick keeps pgrep from matching itself.
WORKER_CMD="python3 /home/super/Projects/NetVM/bin/swarm_worker/daemon.py"
WORKER_MATCH='swarm_worker/daemon.[p]y'

session_exists() {
    tmux -S "$SOCKET_PATH" has-session -t "$SESSION" 2>/dev/null
}

worker_alive() {
    pgrep -f "$WORKER_MATCH" >/dev/null 2>&1
}

do_start() {
    mkdir -p "$LOG_DIR"
    if session_exists; then
        echo "already running (tmux session $SESSION exists on $SOCKET_PATH)"
        return 0
    fi
    tmux -S "$SOCKET_PATH" new-session -d -s "$SESSION" "$WORKER_CMD >>\"$LOG_FILE\" 2>&1"
    sleep 1
    if worker_alive; then
        echo "started tmux session $SESSION on $SOCKET_PATH (logging to $LOG_FILE)"
    else
        echo "WARNING: session $SESSION created on $SOCKET_PATH but worker process not detected yet"
    fi
}

do_stop() {
    local stopped=0
    if session_exists; then
        tmux -S "$SOCKET_PATH" kill-session -t "$SESSION" 2>/dev/null || true
        stopped=1
    fi
    # Also clean up accidental session on default socket if present
    if tmux has-session -t "$SESSION" 2>/dev/null; then
        tmux kill-session -t "$SESSION" 2>/dev/null || true
        stopped=1
    fi
    if [ "$stopped" -eq 1 ]; then
        echo "stopped tmux session $SESSION"
    else
        echo "not running (no tmux session $SESSION)"
    fi
}

do_status() {
    if ! session_exists; then
        echo "STOPPED - no tmux session '$SESSION'"
        return 1
    fi
    if worker_alive; then
        echo "ALIVE - tmux session '$SESSION' exists and worker process is running"
        return 0
    fi
    echo "DEGRADED - tmux session '$SESSION' exists but no worker process found"
    return 2
}

case "${1:-}" in
    start)   do_start ;;
    stop)    do_stop ;;
    restart) do_stop; sleep 1; do_start ;;
    status)  do_status ;;
    *)
        echo "Usage: $(basename "$0") {start|stop|restart|status}" >&2
        exit 2
        ;;
esac