tests/scripts/daemon.sh

#!/usr/bin/env bash
# Start / stop the local reachy-mini daemon cleanly for tests.
#
# WHY: pkill -9 leaks GStreamer semaphores and skips the daemon's
# shutdown procedure (sleep pose, mDNS unregister, peer connection
# teardown). That can leave the local WebRTC pipe in a broken state
# until the laptop reboots — observed 2026-05-06. SIGTERM lets the
# daemon run its proper shutdown.
#
# Usage:
#   tests/scripts/daemon.sh start   # start, wait for "Daemon started successfully"
#   tests/scripts/daemon.sh stop    # SIGTERM, wait for clean exit
#   tests/scripts/daemon.sh restart # stop, then start

set -euo pipefail

LOG=/tmp/reachy-daemon.log
VENV="${VENV:-$HOME/.virtualenvs/mini}"
START_TIMEOUT=60
STOP_TIMEOUT=30

cmd="${1:-}"

start() {
    if pgrep -f reachy-mini-daemon >/dev/null; then
        echo "daemon already running (PID $(pgrep -f reachy-mini-daemon | head -1))"
        return 0
    fi
    # Append (not truncate) so prior sessions' shutdown logs survive for
    # debugging. A separator marks the boundary.
    {
        echo ""
        echo "=== daemon start at $(date -u +%Y-%m-%dT%H:%M:%SZ) ==="
    } >> "$LOG"
    # shellcheck disable=SC1091
    source "$VENV/bin/activate"
    nohup reachy-mini-daemon >> "$LOG" 2>&1 &
    disown
    local pid=$!
    echo "daemon starting (PID $pid)"
    local deadline=$(( $(date +%s) + START_TIMEOUT ))
    # Look ONLY at the lines after the most recent "=== daemon start at"
    # marker. Reset matched=0 on each new marker so we don't trigger on a
    # success line from a previous session that's still in the log.
    until awk '/=== daemon start at /{found=1; matched=0} found && /Daemon started successfully/{matched=1} END{exit !matched}' "$LOG" 2>/dev/null; do
        if ! kill -0 "$pid" 2>/dev/null; then
            echo "daemon died during startup; tail of log:" >&2
            tail -20 "$LOG" >&2
            return 1
        fi
        if [ "$(date +%s)" -gt "$deadline" ]; then
            echo "daemon failed to come up in ${START_TIMEOUT}s" >&2
            tail -20 "$LOG" >&2
            return 1
        fi
        sleep 1
    done
    echo "daemon ready"
}

stop() {
    local pid
    pid=$(pgrep -f reachy-mini-daemon | head -1 || true)
    if [ -z "$pid" ]; then
        echo "no daemon to stop"
        return 0
    fi
    # SIGINT, not SIGTERM. The daemon's full shutdown procedure (put
    # Reachy Mini to sleep, tear down GStreamer pipelines, unregister
    # from the central signaling) is wired to KeyboardInterrupt, which
    # SIGINT raises and SIGTERM does not always trigger reliably across
    # Python versions and entry-point wrappers. Behaves identically to
    # CTRL-C in a foreground shell.
    echo "sending SIGINT to $pid"
    kill -INT "$pid"
    local deadline=$(( $(date +%s) + STOP_TIMEOUT ))
    while kill -0 "$pid" 2>/dev/null; do
        if [ "$(date +%s)" -gt "$deadline" ]; then
            # Last resort. Document the failure loud and proud — if we
            # ever fall through to SIGKILL, semaphores leak and the
            # local WebRTC pipe state is suspect.
            echo "daemon did not exit in ${STOP_TIMEOUT}s after SIGINT; sending SIGKILL (will leak GST resources)" >&2
            kill -KILL "$pid"
            return 1
        fi
        sleep 1
    done
    # Verify the FULL shutdown sequence ran. The daemon walks through:
    #   "Putting Reachy Mini to sleep..."  ← motors safe-pose
    #   "Daemon stopped successfully."     ← clean teardown
    # Both should appear after the latest start marker. If only the
    # second is present, something in the sleep procedure failed and
    # the next session may inherit a half-broken state.
    local sleep_line=0 stopped_line=0
    if awk '/=== daemon start at /{f=1; sleep_seen=0; stopped_seen=0} f && /Putting Reachy Mini to sleep/{sleep_seen=1} f && /Daemon stopped successfully/{stopped_seen=1} END{exit !(sleep_seen && stopped_seen)}' "$LOG" 2>/dev/null; then
        echo "daemon stopped cleanly (sleep pose + full teardown ran)"
    elif awk '/=== daemon start at /{f=1; m=0} f && /Daemon stopped successfully/{m=1} END{exit !m}' "$LOG" 2>/dev/null; then
        echo "WARN: daemon exited but did NOT run sleep pose; next session may be unhealthy" >&2
        tail -15 "$LOG" >&2
    else
        echo "WARN: daemon exited without clean-shutdown markers in log; tail:" >&2
        tail -15 "$LOG" >&2
    fi
}

case "$cmd" in
    start) start ;;
    stop) stop ;;
    restart) stop && start ;;
    *) echo "usage: $0 {start|stop|restart}" >&2; exit 2 ;;
esac