Merge "Fix RabbitMQ ocf_run with the timeout command."
This commit is contained in:
Jenkins
Gerrit Code Review
commit
9b26cb98e7
@ -207,6 +207,25 @@ END
|
||||
#######################################################################
|
||||
# Functions invoked by resource manager actions
|
||||
|
||||
# Invokes the given command as a rabbitmq user and wrapped in the
|
||||
# timeout command.
|
||||
su_rabbit_cmd() {
|
||||
local cmd=${1:-status}
|
||||
local LH="${LL} su_rabbit_cmd():"
|
||||
local rc=1
|
||||
local user=$OCF_RESKEY_username
|
||||
local mail=/var/spool/mail/rabbitmq
|
||||
local pwd=/var/lib/rabbitmq
|
||||
local home=/var/lib/rabbitmq
|
||||
|
||||
ocf_log debug "${LH} invoking a command: ${cmd}"
|
||||
su $user -s /bin/sh -c "USER=${user} MAIL=${mail} PWD=${pwd} HOME=${home} LOGNAME=${user} \
|
||||
${COMMAND_TIMEOUT} ${cmd}"
|
||||
rc=$?
|
||||
ocf_log info "${LH} the invoked command exited ${rc}: ${cmd}"
|
||||
return $rc
|
||||
}
|
||||
|
||||
now() {
|
||||
date -u +%s
|
||||
}
|
||||
@ -323,13 +342,13 @@ reset_mnesia() {
|
||||
if ! $make_amnesia ; then
|
||||
# rabbit app is not running, reset mnesia
|
||||
ocf_log info "${LH} Execute reset with timeout: ${TIMEOUT_ARG}"
|
||||
ocf_run ${COMMAND_TIMEOUT} ${OCF_RESKEY_ctl} reset
|
||||
su_rabbit_cmd "${OCF_RESKEY_ctl} reset"
|
||||
rc=$?
|
||||
if [[ $rc != $OCF_SUCCESS ]] ; then
|
||||
if [[ $rc != 0 ]] ; then
|
||||
ocf_log info "${LH} Execute force_reset with timeout: ${TIMEOUT_ARG}"
|
||||
ocf_run ${COMMAND_TIMEOUT} ${OCF_RESKEY_ctl} force_reset
|
||||
su_rabbit_cmd "${OCF_RESKEY_ctl} force_reset"
|
||||
rc=$?
|
||||
if [[ $rc != $OCF_SUCCESS ]] ; then
|
||||
if [[ $rc != 0 ]] ; then
|
||||
ocf_log warn "${LH} Mnesia couldn't cleaned, even by force-reset command."
|
||||
make_amnesia=true
|
||||
fi
|
||||
@ -504,9 +523,9 @@ join_to_cluster() {
|
||||
fi
|
||||
fi
|
||||
ocf_log info "${LH} Execute join_cluster with timeout: ${TIMEOUT_ARG}"
|
||||
ocf_run ${COMMAND_TIMEOUT} ${OCF_RESKEY_ctl} join_cluster $rmq_node
|
||||
su_rabbit_cmd "${OCF_RESKEY_ctl} join_cluster $rmq_node"
|
||||
rc=$?
|
||||
if [[ $rc != $OCF_SUCCESS ]] ; then
|
||||
if [[ $rc != 0 ]] ; then
|
||||
ocf_log err "${LH} Can't join to cluster by node '${rmq_node}'. Stopping."
|
||||
action_stop
|
||||
return $OCF_ERR_GENERIC
|
||||
@ -566,9 +585,9 @@ unjoin_nodes_from_cluster() {
|
||||
|
||||
# unjoin node
|
||||
ocf_log info "${LH} Execute forget_cluster_node with timeout: ${TIMEOUT_ARG}"
|
||||
ocf_run ${COMMAND_TIMEOUT} ${OCF_RESKEY_ctl} forget_cluster_node ${nodename} 2>&1
|
||||
su_rabbit_cmd "${OCF_RESKEY_ctl} forget_cluster_node ${nodename}"
|
||||
rc=$?
|
||||
if [[ $rc == $OCF_SUCCESS ]] ; then
|
||||
if [[ $rc == 0 ]] ; then
|
||||
ocf_log info "${LH} node '${nodename}' unjoined succesfully."
|
||||
else
|
||||
ocf_log info "${LH} unjoining node '${nodename}' failed."
|
||||
@ -589,8 +608,7 @@ stop_server_process() {
|
||||
rc=$?
|
||||
if [[ $rc != 0 ]] ; then
|
||||
ocf_log err "${LH} RMQ-server process PIDFILE was not found!"
|
||||
#FIXME(bogdando) replace ocf_run to su_rabbit_cmd
|
||||
ocf_run ${COMMAND_TIMEOUT} ${OCF_RESKEY_ctl} stop 2>&1 >> "${OCF_RESKEY_log_dir}/shutdown_log"
|
||||
su_rabbit_cmd "${OCF_RESKEY_ctl} stop 2>&1 >> \"${OCF_RESKEY_log_dir}/shutdown_log\""
|
||||
rc=$?
|
||||
if [[ $rc == 0 ]] ; then
|
||||
ocf_log info "${LH} RMQ-server process stopped succesfully, although there was no PIDFILE found."
|
||||
@ -602,9 +620,9 @@ stop_server_process() {
|
||||
fi
|
||||
|
||||
ocf_log info "${LH} Execute stop with timeout: ${TIMEOUT_ARG}"
|
||||
ocf_run ${COMMAND_TIMEOUT} ${OCF_RESKEY_ctl} stop ${OCF_RESKEY_pid_file} 2>&1 >> "${OCF_RESKEY_log_dir}/shutdown_log"
|
||||
su_rabbit_cmd "${OCF_RESKEY_ctl} stop ${OCF_RESKEY_pid_file} 2>&1 >> \"${OCF_RESKEY_log_dir}/shutdown_log\""
|
||||
rc=$?
|
||||
if [[ $rc == $OCF_SUCCESS ]] ; then
|
||||
if [[ $rc == 0 ]] ; then
|
||||
ocf_log info "${LH} RMQ-server process (PID=${pid}) stopped succesfully."
|
||||
else
|
||||
# RMQ-server process can't stop succesfully
|
||||
@ -632,9 +650,9 @@ stop_rmq_server_app() {
|
||||
|
||||
# stop the app
|
||||
ocf_log info "${LH} Execute stop_app with timeout: ${TIMEOUT_ARG}"
|
||||
ocf_run ${COMMAND_TIMEOUT} ${OCF_RESKEY_ctl} stop_app 2>&1 >> "${OCF_RESKEY_log_dir}/shutdown_log"
|
||||
su_rabbit_cmd "${OCF_RESKEY_ctl} stop_app 2>&1 >> \"${OCF_RESKEY_log_dir}/shutdown_log\""
|
||||
rc=$?
|
||||
if [[ $rc != $OCF_SUCCESS ]] ; then
|
||||
if [[ $rc != 0 ]] ; then
|
||||
ocf_log err "${LH} RMQ-server app cannot be stopped."
|
||||
return $OCF_ERR_GENERIC
|
||||
fi
|
||||
@ -680,7 +698,8 @@ start_beam_process() {
|
||||
[ -f /etc/default/rabbitmq-server ] && . /etc/default/rabbitmq-server
|
||||
|
||||
# run beam process
|
||||
RABBITMQ_NODE_ONLY=1 ${OCF_RESKEY_binary} >> "${OCF_RESKEY_log_dir}/startup_log" 2>/dev/null &
|
||||
local command="${OCF_RESKEY_binary} >> \"${OCF_RESKEY_log_dir}/startup_log\" 2>/dev/null"
|
||||
RABBITMQ_NODE_ONLY=1 su rabbitmq -s /bin/sh -c "${command}"&
|
||||
ts_end=$(( $(now) + ${OCF_RESKEY_start_time} ))
|
||||
rc=$OCF_ERR_GENERIC
|
||||
while [ $(now) -lt ${ts_end} ]; do
|
||||
@ -747,14 +766,14 @@ try_to_start_rmq_app() {
|
||||
|
||||
ocf_log info "${LH} begin."
|
||||
ocf_log info "${LH} Execute start_app with timeout: ${TIMEOUT_ARG}"
|
||||
ocf_run ${COMMAND_TIMEOUT} ${OCF_RESKEY_ctl} start_app >>$startup_log 2>&1
|
||||
su_rabbit_cmd "${OCF_RESKEY_ctl} start_app >>${startup_log} 2>&1"
|
||||
rc=$?
|
||||
if [[ $rc == $OCF_SUCCESS ]] ; then
|
||||
if [[ $rc == 0 ]] ; then
|
||||
ocf_log info "${LH} start_app was successful."
|
||||
ocf_log info "${LH} waiting for start to finish with timeout: ${TIMEOUT_ARG}"
|
||||
ocf_run ${COMMAND_TIMEOUT} ${OCF_RESKEY_ctl} wait ${OCF_RESKEY_pid_file}
|
||||
su_rabbit_cmd "${OCF_RESKEY_ctl} wait ${OCF_RESKEY_pid_file}"
|
||||
rc=$?
|
||||
if [[ $rc != $OCF_SUCCESS ]] ; then
|
||||
if [[ $rc != 0 ]] ; then
|
||||
ocf_log err "${LH} RMQ-server app failed to wait for start."
|
||||
return $OCF_ERR_GENERIC
|
||||
fi
|
||||
|
||||
Loading…
Reference in New Issue
Block a user