openstack/fuel-library
Code Issues Proposed changes

Fix mysql OCF race

Browse Source 
- Fix bug
- Fix issues found in bashate

Closes-Bug: #1658144

Change-Id: I93e9ab269d3ad4a285154fafcac6426ef36f6b52
This commit is contained in:
Sergii Golovatiuk 2017-01-20 20:09:13 +01:00
parent d8269d6f11
commit 2326ef963a
1 changed files with 22 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

41
files/fuel-ha-utils/ocf/mysql-wss
View File

@ -72,9 +72,9 @@ MYSQL=$OCF_RESKEY_client_binary
HOSTNAME=$(uname -n)
MYSQL_OPTIONS_LOCAL="-S $OCF_RESKEY_socket --connect_timeout=10"
if [ "${OCF_RESKEY_test_conf}" ]; then
MYSQL_OPTIONS_TEST="--defaults-extra-file=${OCF_RESKEY_test_conf} ${MYSQL_OPTIONS_LOCAL}"
MYSQL_OPTIONS_TEST="--defaults-extra-file=${OCF_RESKEY_test_conf} ${MYSQL_OPTIONS_LOCAL}"
else
MYSQL_OPTIONS_TEST="$MYSQL_OPTIONS_LOCAL --user=$OCF_RESKEY_test_user --password=$OCF_RESKEY_test_passwd"
MYSQL_OPTIONS_TEST="$MYSQL_OPTIONS_LOCAL --user=$OCF_RESKEY_test_user --password=$OCF_RESKEY_test_passwd"
fi
#######################################################################
usage() {
@ -337,7 +337,8 @@ clear_node_pc()
get_master_timeout() {
local LH="${LL} get_master_timeout():"
local timeout=$(crm_attribute --quiet --name galera_master_timeout \
local timeout
timeout=$(crm_attribute --quiet --name galera_master_timeout \
--query --default=$OCF_RESKEY_master_timeout -q | sed -e '/(null)/d')
ocf_log info "${LH} Setting timeout $timeout"
@ -374,8 +375,8 @@ get_node_gtid_with_retry() {
GTID=$(get_node_gtid $NODE)
if [ "$GTID"="0" ]; then
sleep $[ ( $RANDOM % 10 ) + 1]
if [ "$GTID" = "0" ]; then
sleep $(( ( $RANDOM % 10 ) + 1 ))
GTID=$(get_node_gtid $NODE)
fi
@ -384,12 +385,16 @@ get_node_gtid_with_retry() {
check_if_reelection_needed() {
local LH="${LL} check_if_reelection_needed()"
local PARTITION_WITH_QUORUM=$(crm_node -q | sed -e '/(null)/d')
local RESOURCE_NAME=$(echo $OCF_RESOURCE_INSTANCE | cut -f1 -d":")
local NODE_COUNT=$(nodes_in_cluster | wc -w)
local PARTITION_WITH_QUORUM
local RESOURCE_NAME
local NODE_COUNT
local RUNNING_INSTANCES
local rc
PARTITION_WITH_QUORUM=$(crm_node -q | sed -e '/(null)/d')
RESOURCE_NAME=$(echo $OCF_RESOURCE_INSTANCE | cut -f1 -d":")
NODE_COUNT=$(nodes_in_cluster | wc -w)
if [ $PARTITION_WITH_QUORUM -eq 1 -o $NODE_COUNT -eq 1 ]; then
RUNNING_INSTANCES=$(crm_resource \
--quiet --locate --resource $RESOURCE_NAME | sed -e '/(null)/d' | wc -l 2> /dev/null)
@ -452,8 +457,7 @@ get_master() {
ocf_log info "${LH} The most seen GTID is: ${MASTER_GTID}"
for NODE in $NODES; do
NODE_SCORE=$(crm_simulate -Ls | awk "/${OCF_RESOURCE_INSTANCE}/ && /clone_color/ && ! /${OCF_RESOURCE_INSTANCE}:/ && /${NODE}/ {print \$NF}")
if [[ $NODE_SCORE =~ ^-?[0-9]+$ && $NODE_SCORE -le 0 || $NODE_SCORE = "-INFINITY" || -z $NODE_SCORE ]]
then
if [[ $NODE_SCORE =~ ^-?[0-9]+$ && $NODE_SCORE -le 0 || $NODE_SCORE = "-INFINITY" || -z $NODE_SCORE ]]; then
ocf_log info "${LH} Skipping node $NODE as it is not eligible for running the resource. Its score is ${NODE_SCORE:-NULL}"
continue
fi
@ -487,11 +491,13 @@ check_if_galera_pc() {
local LH="${LL} check_if_galera_pc():"
local NODES
local MASTER
local timeout=$(get_master_timeout)
local timeout
local GTID
local pid
local pcnum=0
timeout=$(get_master_timeout)
ocf_log info "${LH} Checking if Primary Component"
while [ $timeout -gt 0 ]; do
@ -508,15 +514,12 @@ check_if_galera_pc() {
ocf_log info "${LH} My neighbour is Primary Component with GTID: ${GTID}"
if check_if_new_cluster
then
for node in ${NODES}
do
for node in ${NODES}; do
is_pc=$(crm_attribute --quiet --node ${node} --lifetime reboot --query --name is_pc | sed -e '/(null)/d')
if [ ${is_pc} == "true" ]
then
let pcnum=pcnum+1
if [ "${is_pc}" = "true" ]; then
pcnum=$((pcnum + 1))
fi
if [ ${pcnum} -gt 1 ]
then
if [ ${pcnum} -gt 1 ]; then
ocf_log err "${LH} But I'm running a new cluster, PID:${pid}, this is a split-brain!"
exit $OCF_ERR_GENERIC
fi
@ -527,7 +530,7 @@ check_if_galera_pc() {
fi
sleep 10
(( timeout -= 10 ))
timeout=$((timeout - 10))
ocf_log info "${LH} Waiting for master. ${timeout} seconds left"
done