X-Git-Url: https://gerrit.opnfv.org/gerrit/gitweb?a=blobdiff_plain;f=extraconfig%2Ftasks%2Fmajor_upgrade_controller_pacemaker_2.sh;h=8b900842c0c68b10a9e7b9ba0b3ca42fd681ba77;hb=864cb733a3b11f62a79d724f979d672c234d8986;hp=4203eba9e6ae82985f03a53194a1d03cf028a85b;hpb=1bf2b3cc0b7c07a183fee4488498a863a5634ed9;p=apex-tripleo-heat-templates.git diff --git a/extraconfig/tasks/major_upgrade_controller_pacemaker_2.sh b/extraconfig/tasks/major_upgrade_controller_pacemaker_2.sh index 4203eba9..8b900842 100755 --- a/extraconfig/tasks/major_upgrade_controller_pacemaker_2.sh +++ b/extraconfig/tasks/major_upgrade_controller_pacemaker_2.sh @@ -2,69 +2,176 @@ set -eu -cluster_form_timeout=600 -cluster_settle_timeout=1800 -galera_sync_timeout=600 +cluster_sync_timeout=1800 -if [[ -n $(is_bootstrap_node) ]]; then - pcs cluster start --all +# After migrating the cluster to HA-NG the services not under pacemaker's control +# are still up and running. We need to stop them explicitely otherwise during the yum +# upgrade the rpm %post sections will try to do a systemctl try-restart , which +# is going to take a long time because rabbit is down. By having the service stopped +# systemctl try-restart is a noop +for service in $(services_to_migrate); do + manage_systemd_service stop "${service%%-clone}" + # So the reason for not reusing check_resource_systemd is that + # I have observed systemctl is-active returning unknown with at least + # one service that was stopped (See LP 1627254) + timeout=600 tstart=$(date +%s) - while pcs status 2>&1 | grep -E '(cluster is not currently running)|(OFFLINE:)'; do - sleep 5 - tnow=$(date +%s) - if (( tnow-tstart > cluster_form_timeout )) ; then - echo_error "ERROR: timed out forming the cluster" - exit 1 - fi + tend=$(( $tstart + $timeout )) + check_interval=3 + while (( $(date +%s) < $tend )); do + if [[ "$(systemctl is-active ${service%%-clone})" = "active" ]]; then + echo "$service still active, sleeping $check_interval seconds." + sleep $check_interval + else + # we do not care if it is inactive, unknown or failed as long as it is + # not running + break + fi + done +done - if ! timeout -k 10 $cluster_settle_timeout crm_resource --wait; then - echo_error "ERROR: timed out waiting for cluster to finish transition" - exit 1 +# In case the mysql package is updated, the database on disk must be +# upgraded as well. This typically needs to happen during major +# version upgrades (e.g. 5.5 -> 5.6, 5.5 -> 10.1...) +# +# Because in-place upgrades are not supported across 2+ major versions +# (e.g. 5.5 -> 10.1), we rely on logical upgrades via dump/restore cycle +# https://bugzilla.redhat.com/show_bug.cgi?id=1341968 +# +# The default is to determine automatically if upgrade is needed based +# on mysql package versionning, but this can be overriden manually +# to support specific upgrade scenario + +# Calling this function will set the DO_MYSQL_UPGRADE variable which is used +# later +mysql_need_update + +if [[ -n $(is_bootstrap_node) ]]; then + if [ $DO_MYSQL_UPGRADE -eq 1 ]; then + backup_flags="--defaults-extra-file=/root/.my.cnf -u root --flush-privileges --all-databases --single-transaction" + mysqldump $backup_flags > "$MYSQL_BACKUP_DIR/openstack_database.sql" + cp -rdp /etc/my.cnf* "$MYSQL_BACKUP_DIR" fi - for vip in $(pcs resource show | grep ocf::heartbeat:IPaddr2 | grep Stopped | awk '{ print $1 }'); do - pcs resource enable $vip - check_resource_pacemaker $vip started 60 + pcs resource disable redis + check_resource redis stopped 600 + pcs resource disable rabbitmq + check_resource rabbitmq stopped 600 + pcs resource disable galera + check_resource galera stopped 600 + pcs resource disable openstack-cinder-volume + check_resource openstack-cinder-volume stopped 600 + # Disable all VIPs before stopping the cluster, so that pcs doesn't use one as a source address: + # https://bugzilla.redhat.com/show_bug.cgi?id=1330688 + for vip in $(pcs resource show | grep ocf::heartbeat:IPaddr2 | grep Started | awk '{ print $1 }'); do + pcs resource disable $vip + check_resource $vip stopped 60 done + pcs cluster stop --all fi -start_or_enable_service galera -check_resource galera started 600 -start_or_enable_service redis -check_resource redis started 600 -# We need mongod which is now a systemd service up and running before calling -# ceilometer-dbsync. There is still a race here: mongod might not be up on all nodes -# so ceilometer-dbsync will fail a couple of times before that. As it retries indefinitely -# we should be good. -# Due to LP Bug https://bugs.launchpad.net/tripleo/+bug/1627254 am using systemctl directly atm -systemctl start mongod -check_resource mongod started 600 -if [[ -n $(is_bootstrap_node) ]]; then - tstart=$(date +%s) - while ! clustercheck; do - sleep 5 - tnow=$(date +%s) - if (( tnow-tstart > galera_sync_timeout )) ; then - echo_error "ERROR galera sync timed out" - exit 1 - fi - done +# Swift isn't controlled by pacemaker +systemctl_swift stop + +tstart=$(date +%s) +while systemctl is-active pacemaker; do + sleep 5 + tnow=$(date +%s) + if (( tnow-tstart > cluster_sync_timeout )) ; then + echo_error "ERROR: cluster shutdown timed out" + exit 1 + fi +done + +# The reason we do an sql dump *and* we move the old dir out of +# the way is because it gives us an extra level of safety in case +# something goes wrong during the upgrade. Once the restore is +# successful we go ahead and remove it. If the directory exists +# we bail out as it means the upgrade process had issues in the last +# run. +if [ $DO_MYSQL_UPGRADE -eq 1 ]; then + if [ -d $MYSQL_TEMP_UPGRADE_BACKUP_DIR ]; then + echo_error "ERROR: mysql backup dir already exist" + exit 1 + fi + mv /var/lib/mysql $MYSQL_TEMP_UPGRADE_BACKUP_DIR +fi - # Run all the db syncs - # TODO: check if this can be triggered in puppet and removed from here - ceilometer-dbsync --config-file=/etc/ceilometer/ceilometer.conf - cinder-manage db sync - glance-manage --config-file=/etc/glance/glance-registry.conf db_sync - heat-manage --config-file /etc/heat/heat.conf db_sync - keystone-manage db_sync - neutron-db-manage --config-file /etc/neutron/neutron.conf --config-file /etc/neutron/plugin.ini upgrade head - nova-manage db sync - nova-manage api_db sync - nova-manage db online_data_migrations - gnocchi-upgrade - #TODO(marios):someone from sahara needs to check this: - # sahara-db-manage --config-file /etc/sahara/sahara.conf upgrade head +# Special-case OVS for https://bugs.launchpad.net/tripleo/+bug/1635205 +special_case_ovs_upgrade_if_needed + +yum -y install python-zaqarclient # needed for os-collect-config +yum -y -q update + +# We need to ensure at least those two configuration settings, otherwise +# mariadb 10.1+ won't activate galera replication. +# wsrep_cluster_address must only be set though, its value does not +# matter because it's overriden by the galera resource agent. +cat >> /etc/my.cnf.d/galera.cnf < /dev/null; do sleep 1; done' + mysql -u root < "$MYSQL_BACKUP_DIR/openstack_database.sql" + mv /root/.my.cnf.temporary /root/.my.cnf + mysqladmin -u root shutdown + # The import was successful so we may remove the folder + rm -r "$MYSQL_BACKUP_DIR" + fi +fi + +# If we reached here without error we can safely blow away the origin +# mysql dir from every controller + +# TODO: What if the upgrade fails on the bootstrap node, but not on +# this controller. Data may be lost. +if [ $DO_MYSQL_UPGRADE -eq 1 ]; then + rm -r $MYSQL_TEMP_UPGRADE_BACKUP_DIR +fi + +# Let's reset the stonith back to true if it was true, before starting the cluster +if [[ -n $(is_bootstrap_node) ]]; then + if [ -f /var/tmp/stonith-true ]; then + pcs -f /var/lib/pacemaker/cib/cib.xml property set stonith-enabled=true + fi + rm -f /var/tmp/stonith-true fi + +# Pin messages sent to compute nodes to kilo, these will be upgraded later +crudini --set /etc/nova/nova.conf upgrade_levels compute "$upgrade_level_nova_compute" +# https://bugzilla.redhat.com/show_bug.cgi?id=1284047 +# Change-Id: Ib3f6c12ff5471e1f017f28b16b1e6496a4a4b435 +crudini --set /etc/ceilometer/ceilometer.conf DEFAULT rpc_backend rabbit +# https://bugzilla.redhat.com/show_bug.cgi?id=1284058 +# Ifd1861e3df46fad0e44ff9b5cbd58711bbc87c97 Swift Ceilometer middleware no longer exists +crudini --set /etc/swift/proxy-server.conf pipeline:main pipeline "catch_errors healthcheck cache ratelimit tempurl formpost authtoken keystone staticweb proxy-logging proxy-server" +# LP: 1615035, required only for M/N upgrade. +crudini --set /etc/nova/nova.conf DEFAULT scheduler_host_manager host_manager +# LP: 1627450, required only for M/N upgrade +crudini --set /etc/nova/nova.conf DEFAULT scheduler_driver filter_scheduler + +crudini --set /etc/sahara/sahara.conf DEFAULT plugins ambari,cdh,mapr,vanilla,spark,storm +