--- ldirectord.orij 2009-11-02 08:40:44.108418595 +1000 +++ ldirectord 2009-11-02 08:45:11.441558547 +1000 @@ -48,16 +48,14 @@ . ${OCF_ROOT}/resource.d/heartbeat/.ocf-shellfuncs -####################################################################### - -# LDIRECTORD=/usr/sbin/ldirectord -# LDIRCONF=/etc/ldirectord.cf +LDIRCONF=${OCF_RESKEY_configfile:-/etc/ha.d/ldirectord.cf} +LDIRECTORD=${OCF_RESKEY_ldirectord:-/usr/sbin/ldirectord} meta_data() { cat < - + 1.0 @@ -101,41 +99,17 @@ ####################################################################### ldir_init() { - - COMMAND=$1 - - LDIRCONF=$OCF_RESKEY_configfile - if [ x"${LDIRCONF}" = "x" ]; then - LDIRCONF=/etc/ldirectord.cf - fi - - if [ ! -f $LDIRCONF ]; then - case $COMMAND in - stop) ocf_log warn "$LDIRCONF not found. ldirectord considered stopped" - exit $OCF_SUCCESS;; - monitior) exit $OCF_NOT_RUNNING;; - status) exit $LSB_STATUS_STOPPED;; - start) ocf_log warn "$LDIRCONF not found." - exit $OCF_NOT_RUNNING;; - esac - fi - - LDIRECTORD=$OCF_RESKEY_ldirectord - if [ x"${LDIRECTORD}" = x ]; then - LDIRECTORD="/usr/sbin/ldirectord" - fi - - if [ ! -x $LDIRECTORD ]; then - case $COMMAND in - stop) ocf_log warn "$LDIRECTORD not found." - exit $OCF_SUCCESS;; - monitor) exit $OCF_NOT_RUNNING;; - status) exit $LSB_STATUS_STOPPED;; - start) ocf_log warn "$LDIRECTORD not found." - exit $OCF_NOT_RUNNING;; - esac - fi - + # check the supplied parameters exist enough that we can do all the other + # operations + if [ ! -f $LDIRCONF ]; then + ocf_log warn "$LDIRCONF not found, ldirectord not installed" + exit $OCF_ERR_INSTALLED + fi + + if [ ! -x $LDIRECTORD ]; then + ocf_log warn "$LDIRECTORD not found, ldirectord not installed" + exit $OCF_ERR_INSTALLED + fi } ldirectord_usage() { @@ -151,58 +125,118 @@ } ldirectord_start() { + ldirectord_status + RET=$? + + # if ldirectord is running or there's an error, pass on this return code + if [ $RET -ne $OCF_NOT_RUNNING ]; then + return $RET + fi + + ocf_log info "Starting ldirectord" + echo $LDIRECTORD $LDIRCONF start $LDIRECTORD $LDIRCONF start + RET=$? + if [ $RET -ne 0 ]; then + return $OCF_ERR_GENERIC + fi + + # call status again to make sure we're running properly + ldirectord_status } ldirectord_stop() { - $LDIRECTORD $LDIRCONF stop + # a status check is simple enough to see if ldirectord is running or not. + # It will also error out if there's configuration parsing errors so we + # can try kill ldirectord even if the config is broken. + ldirectord_status + RET=$? + + # ldirectord may be running, hard to tell when status returns an error + if [ $RET -eq $OCF_ERR_GENERIC ]; then + # get the PID of the right ldirectord process + PID=`pgrep -f "$LDIRECTORD $LDIRCONF start" 2>&1` + RET=$? + + if [ $RET -eq 0 ]; then + ocf_log warn "Killing ldirectord($PID) with SIGTERM" + kill $PID + fi + + pgrep -f "$LDIRECTORD $LDIRCONF start" >/dev/null 2>&1 + RET=$? + # if ldirectord is not running any more, we've (kind of) successfully + # stopped it + if [ $RET -eq 1 ]; then + return $OCF_SUCCESS + else + # ldirectord is still running? Kill it badly + ocf_log warn "Killing ldirectord($PID) with SIGKILL" + kill -9 $PID + + pgrep -f "$LDIRECTORD $LDIRCONF start" >/dev/null 2>&1 + RET=$? + # if it's not dead after here, we can't really do anything more + if [ $RET -eq 1 ]; then + return $OCF_SUCCESS + fi + fi + + # if none of our kills work, return an error. This should force the + # resource unmanaged + # on this node, requiring manual intervention. + return $OCF_ERR_GENERIC + else + ocf_log info "Stopping ldirectord" + # if ldirectord status is not an error, issue a stop. Multiple stops + # will return 0 + $LDIRECTORD $LDIRCONF stop + RET=$? + case $RET in + 0) return $RET;; + *) return 1;; + esac + fi } +# simple check to see if ldirectord is running, returns the proper OCF codes. ldirectord_status() { OUTPUT=`$LDIRECTORD $LDIRCONF status 2>&1` case $? in - 1) echo $OUTPUT - return $OCF_ERR_GENERIC - ;; - 0) echo running - return $OCF_SUCCESS - ;; - 3) echo stopped - return $OCF_SUCCESS - ;; - *) echo $OUTPUT - return $OCF_ERR_GENERIC - ;; + 0) return $OCF_SUCCESS;; + 1) expr match "$OUTPUT" '.*ldirectord stale pid file.*' >/dev/null + if [ $? -eq 0 ]; then + return $OCF_NOT_RUNNING + else + return $OCF_ERR_GENERIC + fi;; + 2) ocf_log err "$LDIRCONF has configuration errors" + echo $OUTPUT + return $OCF_ERR_GENERIC;; + 3) return $OCF_NOT_RUNNING;; + *) echo $OUTPUT + return $OCF_ERR_GENERIC;; esac } ldirectord_monitor() { - OUTPUT=`$LDIRECTORD $LDIRCONF status 2>&1` - case $? in - 0) return $OCF_SUCCESS - ;; - 3) return $OCF_NOT_RUNNING - ;; - *) echo $OUTPUT - return $OCF_ERR_GENERIC - ;; - esac + # check if the process is running first + ldirectord_status + RET=$? + + if [ $RET -ne $OCF_SUCCESS ]; then + return $RET + fi + + # do more advanced checks here for high OCF_CHECK_LEVELs. Don't know what + # more we can do at this time, + # a status call already hits LVS in the kernel. } ldirectord_validate() { - if [ ! -f $LDIRCONF ]; then - ocf_log err $LDIRCONF - ocf_log err "Configuration file $LDIRCONF not found!" - exit $OCF_ERR_CONFIGURED - fi - - if [ ! -x $LDIRECTORD ]; then - ocf_log err "Binary file $LDIRECTORD not found." - exit $OCF_ERR_CONFIGURED - fi - -# exit $OC_ERR_UNIMPLEMENTED - + #ldir_init is already called, there's nothing more we can validate unless + #we add more attributes + return 0 } ldir_init @@ -217,9 +251,6 @@ stop) ldirectord_stop ldirectord_exit $? ;; -status) ldirectord_status - ldirectord_exit $? - ;; monitor) ldirectord_monitor ldirectord_exit $? ;;