diff options
Diffstat (limited to 'sysutils/slurm-wlm/files/slurmctld.in')
| -rw-r--r-- | sysutils/slurm-wlm/files/slurmctld.in | 133 |
1 files changed, 115 insertions, 18 deletions
diff --git a/sysutils/slurm-wlm/files/slurmctld.in b/sysutils/slurm-wlm/files/slurmctld.in index 5199e48b23bc..94f5429687d4 100644 --- a/sysutils/slurm-wlm/files/slurmctld.in +++ b/sysutils/slurm-wlm/files/slurmctld.in @@ -8,36 +8,133 @@ # Add the following lines to /etc/rc.conf.local or /etc/rc.conf # to enable this service: # -# slurmctld_enable (bool): Set to NO by default. -# Set it to YES to enable slurmctld. +# slurmctld_enable (bool): Set to NO by default. +# Set it to YES to enable slurmctld. +# +# Common knobs (honoured by slurmctld and slurmd): +# slurm_user (str): User to run Slurm daemons as (default: slurm) +# slurm_group (str): Group to run Slurm daemons as (default: slurm) +# slurm_conf (str): Path to slurm.conf, exported as SLURM_CONF +# (default: %%ETCDIR%%/slurm.conf) +# slurm_logdir (str): Log directory (default: /var/log/slurm) +# slurm_rundir (str): Runtime directory (default: /var/run/slurm) +# +# Service-specific knobs: +# slurmctld_flags (str): Extra arguments passed to slurmctld. +# slurmctld_pidfile (str): PID file path +# (default: ${slurm_rundir}/slurmctld.pid) +# slurmctld_logfile (str): Log file path +# (default: ${slurm_logdir}/slurmctld.log) # . /etc/rc.subr name="slurmctld" -rcvar=slurmctld_enable - -pidfile=/var/run/$name.pid +rcvar="slurmctld_enable" load_rc_config $name -: ${slurmctld_enable="NO"} +# Common defaults (shared conceptual contract with slurmd) +: ${slurm_user:="slurm"} +: ${slurm_group:="slurm"} +: ${slurm_conf:="%%ETCDIR%%/slurm.conf"} +: ${slurm_logdir:="/var/log/slurm"} +: ${slurm_rundir:="/var/run/slurm"} + +# Service defaults +: ${slurmctld_enable:="NO"} +: ${slurmctld_flags:=""} +: ${slurmctld_pidfile:="${slurm_rundir}/slurmctld.pid"} +: ${slurmctld_logfile:="${slurm_logdir}/slurmctld.log"} + +pidfile="${slurmctld_pidfile}" -start_cmd=slurmctld_start -stop_cmd=slurmctld_stop +# We run slurmctld under daemon(8) for pidfile/log management. +command="/usr/sbin/daemon" +procname="%%PREFIX%%/sbin/${name}" +command_args="-P ${pidfile} -o ${slurmctld_logfile} ${procname} -D ${slurmctld_flags}" + +extra_commands="reload" +start_precmd="${name}_prestart" +reload_cmd="${name}_reload" +status_cmd="${name}_status" +stop_cmd="${name}_stop" + +slurmctld_prestart() +{ + # Ensure log and run directories exist with correct ownership/modes. + /usr/bin/install -d -o "${slurm_user}" -g "${slurm_group}" -m 0750 "${slurm_logdir}" || return 1 + /usr/bin/install -d -o "root" -g "wheel" -m 0755 "${slurm_rundir}" || return 1 + + # Export SLURM_CONF if not already provided in the environment. + if [ -z "${SLURM_CONF}" ]; then + export SLURM_CONF="${slurm_conf}" + fi +} -slurmctld_start() { - checkyesno slurmctld_enable && echo "Starting $name." && \ - %%PREFIX%%/sbin/$name $slurmctld_flags +slurmctld_reload() +{ + if [ ! -r "${pidfile}" ]; then + echo "${name} not running? (pidfile not found)" + return 1 + fi + echo "Reloading ${name} configuration." + kill -HUP "$(cat "${pidfile}")" } -slurmctld_stop() { - if [ -e $pidfile ]; then - checkyesno slurmctld_enable && echo "Stopping $name." && \ - kill `cat $pidfile` - else - killall $name - fi +slurmctld_status() +{ + if [ ! -r "${pidfile}" ]; then + echo "${name} is not running (no pidfile)." + return 1 + fi + if ! check_pidfile "${pidfile}" "${procname}"; then + echo "${name} is not running (stale pidfile)." + return 1 + fi + echo "${name} is running as pid $(cat "${pidfile}")." +} + +slurmctld_stop() +{ + if [ ! -r "${pidfile}" ]; then + echo "${name} not running? (no pidfile)." + return 1 + fi + + if ! check_pidfile "${pidfile}" "${procname}"; then + echo "${name} not running? (stale pidfile: ${pidfile})." + rm -f "${pidfile}" + return 1 + fi + + pid="$(cat "${pidfile}" 2>/dev/null || true)" + if [ -z "${pid}" ]; then + echo "${name} not running? (empty pidfile: ${pidfile})." + rm -f "${pidfile}" + return 1 + fi + + echo "Stopping ${name} (pid ${pid})." + kill -TERM "${pid}" 2>/dev/null || true + + # Allow a short grace period for clean shutdown and helper cleanup. + i=0 + while [ $i -lt 5 ]; do + if ! check_pidfile "${pidfile}" "${procname}"; then + rm -f "${pidfile}" + return 0 + fi + i=$((i + 1)) + sleep 1 + done + + echo "${name} did not exit on SIGTERM; sending SIGKILL to controller and its helpers." + # Kill children (e.g. slurmscriptd) first, then the controller itself. + pkill -KILL -P "${pid}" 2>/dev/null || true + kill -KILL "${pid}" 2>/dev/null || true + + rm -f "${pidfile}" } run_rc_command "$1" |