Merge pull request 'fix(rc.d): supervisor-aware stop + bring clawdie.in to parity (Sam & Claude)' (#18) from fix/rc.d-stop-supervisor into main

Reviewed-on: #18

packaging/freebsd/clawdie.in

@@ -9,9 +9,8 @@
 # /usr/local/etc/clawdie/clawdie.env) or set them in rc.conf below.
 #
 # clawdie runs in the FOREGROUND (no self-daemonize, no pidfile), so rc.d runs
-# it under daemon(8), which backgrounds it, writes the supervisor pidfile,
+# it under daemon(8), which backgrounds it, restarts on crash, drops privileges
-# restarts on crash, drops privileges to the clawdie user, and redirects
+# to the clawdie user, and redirects stdout/stderr (tracing) to a logfile.
-# stdout/stderr (tracing) to a logfile.
 #
 # Install:
 #   pw groupadd clawdie
@@ -47,18 +46,29 @@ load_rc_config $name
 : ${clawdie_env_file:="/usr/local/etc/clawdie/clawdie.env"}
 
 pidfile="${clawdie_run_dir}/clawdie.pid"
+# Supervisor pidfile (the daemon(8) parent). Kept distinct from the child
+# pidfile so `stop` can target the supervisor — see clawdie_stop.
+supervisor_pidfile="${clawdie_run_dir}/clawdie-supervisor.pid"
 
-# Supervise via daemon(8): -P supervisor pidfile, -r restart on exit, -t title,
+# Run clawdie under daemon(8):
-# -u drop privileges, -o append stdout/stderr to the logfile.
+#   -P supervisor pidfile (the daemon(8) parent — used by stop)
+#   -p child pidfile (writes the clawdie PID — used by start/status)
+#   -r restart on crash, -t process title, -u drop to the clawdie user,
+#   -o append stdout/stderr to log.
 command="/usr/sbin/daemon"
-command_args="-P ${pidfile} -r -t ${name} -u ${clawdie_user} \
+command_args="-P ${supervisor_pidfile} -p ${pidfile} -r -t ${name} -u ${clawdie_user} \
     -o ${clawdie_logfile} ${clawdie_program}"
 
-# rc.subr matches the pidfile process against ${procname}; that is daemon(8),
+# Match the child binary so `service clawdie status` finds OUR process via the
-# the supervised parent — not clawdie itself.
+# child pidfile, not the generic /usr/sbin/daemon supervisor (which would
-procname="/usr/sbin/daemon"
+# collide with tailscaled, colibri_daemon, and other daemon(8) services).
+procname="clawdie"
 
 start_precmd="clawdie_prestart"
+start_postcmd="clawdie_poststart"
+stop_cmd="clawdie_stop"
+stop_postcmd="clawdie_poststop"
+extra_commands="health"
 
 clawdie_prestart()
 {
@@ -89,4 +99,79 @@ clawdie_prestart()
     fi
 }
 
+clawdie_poststart()
+{
+    # Wait for the Herdr socket to appear (daemon forks, child binds socket).
+    local timeout=10
+    local waited=0
+    while [ ! -S "${clawdie_socket}" ] && [ $waited -lt $timeout ]; do
+        sleep 1
+        waited=$((waited + 1))
+    done
+
+    if [ -S "${clawdie_socket}" ]; then
+        echo "clawdie socket ready after ${waited}s"
+    else
+        echo "WARNING: clawdie socket not ready after ${timeout}s"
+    fi
+}
+
+clawdie_stop()
+{
+    # daemon(8) -r restarts the child if it is killed directly, so a plain
+    # SIGTERM to the child pidfile would just be undone. Stop the supervisor
+    # instead: on SIGTERM it forwards the signal to the child and exits without
+    # restarting it.
+    local _sup=""
+    [ -f "${supervisor_pidfile}" ] && _sup=$(cat "${supervisor_pidfile}" 2>/dev/null)
+    if [ -n "${_sup}" ] && kill -0 "${_sup}" 2>/dev/null; then
+        echo "Stopping ${name} (daemon(8) supervisor pid ${_sup})."
+        kill -TERM "${_sup}" 2>/dev/null
+        local _n=0
+        while kill -0 "${_sup}" 2>/dev/null && [ ${_n} -lt 30 ]; do
+            sleep 1
+            _n=$((_n + 1))
+        done
+        if kill -0 "${_sup}" 2>/dev/null; then
+            echo "Supervisor did not exit in time; sending SIGKILL."
+            kill -KILL "${_sup}" 2>/dev/null
+        fi
+    else
+        echo "${name} is not running."
+    fi
+    # Belt-and-suspenders: terminate the child if it somehow outlived the
+    # supervisor (e.g. supervisor SIGKILLed before it could clean up).
+    local _ch=""
+    [ -f "${pidfile}" ] && _ch=$(cat "${pidfile}" 2>/dev/null)
+    if [ -n "${_ch}" ] && kill -0 "${_ch}" 2>/dev/null; then
+        kill -TERM "${_ch}" 2>/dev/null
+    fi
+    rm -f "${supervisor_pidfile}" "${pidfile}"
+}
+
+clawdie_poststop()
+{
+    # Clean up tmpfs artifacts on graceful shutdown.
+    if [ -S "${clawdie_socket}" ]; then
+        rm -f "${clawdie_socket}"
+    fi
+}
+
+health_cmd="clawdie_health"
+clawdie_health()
+{
+    if [ -S "${clawdie_socket}" ]; then
+        if printf '{"cmd":"status"}\n' | nc -U "${clawdie_socket}" -w 2 >/dev/null 2>&1; then
+            echo "clawdie is healthy (socket responding)"
+            return 0
+        else
+            echo "clawdie socket exists but not responding"
+            return 1
+        fi
+    else
+        echo "clawdie socket not found"
+        return 1
+    fi
+}
+
 run_rc_command "$1"

packaging/freebsd/colibri_daemon.in

@@ -47,12 +47,17 @@ load_rc_config $name
 : ${colibri_cost_mode:="smart"}
 
 pidfile="${colibri_daemon_run_dir}/colibri-daemon.pid"
+# Supervisor pidfile (the daemon(8) parent). Kept distinct from the child
+# pidfile so `stop` can target the supervisor — see colibri_daemon_stop.
+supervisor_pidfile="${colibri_daemon_run_dir}/colibri-daemon-supervisor.pid"
 
-# Run colibri-daemon under daemon(8): -p child pidfile (writes colibri-daemon PID),
+# Run colibri-daemon under daemon(8):
-# -r restart on exit, -t process title, -u drop to the colibri user, -o append
+#   -P supervisor pidfile (the daemon(8) parent — used by stop)
-# stdout/stderr to log.
+#   -p child pidfile (writes colibri-daemon PID — used by start/status)
+#   -r restart on crash, -t process title, -u drop to the colibri user,
+#   -o append stdout/stderr to log.
 command="/usr/sbin/daemon"
-command_args="-p ${pidfile} -r -t ${name} -u ${colibri_daemon_user} \
+command_args="-P ${supervisor_pidfile} -p ${pidfile} -r -t ${name} -u ${colibri_daemon_user} \
     -o ${colibri_daemon_logfile} ${colibri_daemon_program}"
 
 # Use the child's process name so rc.subr can find the right process via the
@@ -62,6 +67,7 @@ procname="colibri-daemon"
 
 start_precmd="colibri_daemon_prestart"
 start_postcmd="colibri_daemon_poststart"
+stop_cmd="colibri_daemon_stop"
 stop_postcmd="colibri_daemon_poststop"
 extra_commands="health"
 
@@ -100,6 +106,39 @@ colibri_daemon_poststart()
     fi
 }
 
+colibri_daemon_stop()
+{
+    # daemon(8) -r restarts the child if it is killed directly, so a plain
+    # SIGTERM to the child pidfile would just be undone. Stop the supervisor
+    # instead: on SIGTERM it forwards the signal to the child and exits without
+    # restarting it.
+    local _sup=""
+    [ -f "${supervisor_pidfile}" ] && _sup=$(cat "${supervisor_pidfile}" 2>/dev/null)
+    if [ -n "${_sup}" ] && kill -0 "${_sup}" 2>/dev/null; then
+        echo "Stopping ${name} (daemon(8) supervisor pid ${_sup})."
+        kill -TERM "${_sup}" 2>/dev/null
+        local _n=0
+        while kill -0 "${_sup}" 2>/dev/null && [ ${_n} -lt 30 ]; do
+            sleep 1
+            _n=$((_n + 1))
+        done
+        if kill -0 "${_sup}" 2>/dev/null; then
+            echo "Supervisor did not exit in time; sending SIGKILL."
+            kill -KILL "${_sup}" 2>/dev/null
+        fi
+    else
+        echo "${name} is not running."
+    fi
+    # Belt-and-suspenders: terminate the child if it somehow outlived the
+    # supervisor (e.g. supervisor SIGKILLed before it could clean up).
+    local _ch=""
+    [ -f "${pidfile}" ] && _ch=$(cat "${pidfile}" 2>/dev/null)
+    if [ -n "${_ch}" ] && kill -0 "${_ch}" 2>/dev/null; then
+        kill -TERM "${_ch}" 2>/dev/null
+    fi
+    rm -f "${supervisor_pidfile}" "${pidfile}"
+}
+
 colibri_daemon_poststop()
 {
     # Clean up tmpfs artifacts on graceful shutdown.