From 63421ca73de8ed9a0033706c190bf9a6eca4b582 Mon Sep 17 00:00:00 2001 From: Michele Baldessari Date: Fri, 26 Aug 2016 16:46:44 +0200 Subject: Add a function to upgrade from full HA to NG HA This is the initial work to have a function that migrates a full HA architecture as deployed in Mitaka to the HA architecture as deployed in Newton where only a few resources are managed by pacemaker. The sequence is the following: 1) We remove the desired services from pacemaker's control. The services at this point are still running normally via the systemd service as invoked by pacemaker 2) We do a "systemctl stop " on all controllers for all the services that were removed from pacemaker's control. We do this to make sure that during the yum upgrade, the %post sections that call "systemctl try-restart" do not take ages, because at this point during the upgrade rabbit is down. The only exceptions are "openstack-core" and "delay" which are dummy pacemaker resources that do not exist on the system 3) We do a "systemctl start " on all nodes for all the services mentioned above. We should probably merge this patch only when newton has branched as it is very specific to the M/N upgrade. Closes-Bug: 1617520 Change-Id: I4c409ce58c1a57b6e0decc3cf168b62698b32e39 --- .../tasks/major_upgrade_controller_pacemaker_1.sh | 26 +++-- .../tasks/major_upgrade_controller_pacemaker_2.sh | 14 +-- .../tasks/major_upgrade_pacemaker_migrations.sh | 113 +++++++++++++++++++++ 3 files changed, 137 insertions(+), 16 deletions(-) (limited to 'extraconfig') diff --git a/extraconfig/tasks/major_upgrade_controller_pacemaker_1.sh b/extraconfig/tasks/major_upgrade_controller_pacemaker_1.sh index 6d073d86..2490ce27 100755 --- a/extraconfig/tasks/major_upgrade_controller_pacemaker_1.sh +++ b/extraconfig/tasks/major_upgrade_controller_pacemaker_1.sh @@ -18,6 +18,22 @@ check_disk_for_mysql_dump STONITH_STATE=$(pcs property show stonith-enabled | grep "stonith-enabled" | awk '{ print $2 }') pcs property set stonith-enabled=false +# Migrate to HA NG +if [ "$(hiera -c /etc/puppet/hiera.yaml bootstrap_nodeid)" = "$(facter hostname)" ]; then + migrate_full_to_ng_ha +fi + +# 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}" + check_resource_systemd "${service%%-clone}" stopped 600 +done + # In case the mysql package is updated, the database on disk must be # upgraded as well. This typically needs to happen during major # version upgrades (e.g. 5.5 -> 5.6, 5.5 -> 10.1...) @@ -36,8 +52,6 @@ if [ "$(hiera -c /etc/puppet/hiera.yaml bootstrap_nodeid)" = "$(facter hostname) cp -rdp /etc/my.cnf* "$MYSQL_BACKUP_DIR" fi - pcs resource disable httpd - check_resource httpd stopped 1800 pcs resource disable redis check_resource redis stopped 600 pcs resource disable rabbitmq @@ -53,14 +67,6 @@ if [ "$(hiera -c /etc/puppet/hiera.yaml bootstrap_nodeid)" = "$(facter hostname) pcs cluster stop --all fi -stop_or_disable_service mongod -check_resource mongod stopped 600 -stop_or_disable_service memcached -check_resource memcached stopped 600 - - - - # Swift isn't controled by pacemaker systemctl_swift stop diff --git a/extraconfig/tasks/major_upgrade_controller_pacemaker_2.sh b/extraconfig/tasks/major_upgrade_controller_pacemaker_2.sh index cfe5bcfe..6bb2fa73 100755 --- a/extraconfig/tasks/major_upgrade_controller_pacemaker_2.sh +++ b/extraconfig/tasks/major_upgrade_controller_pacemaker_2.sh @@ -32,8 +32,6 @@ fi start_or_enable_service galera check_resource galera started 600 -start_or_enable_service mongod -check_resource mongod started 600 if [[ -n $(is_bootstrap_node) ]]; then tstart=$(date +%s) @@ -59,14 +57,18 @@ if [[ -n $(is_bootstrap_node) ]]; then # sahara-db-manage --config-file /etc/sahara/sahara.conf upgrade head fi -start_or_enable_service memcached -check_resource memcached started 600 start_or_enable_service rabbitmq check_resource rabbitmq started 600 start_or_enable_service redis check_resource redis started 600 -start_or_enable_service httpd -check_resource httpd started 1800 # Swift isn't controled by pacemaker systemctl_swift start + +# We need to start the systemd services we explicitely stopped at step _1.sh +# FIXME: Should we let puppet during the convergence step do the service enabling or +# should we add it here? +for $service in $(services_to_migrate); do + manage_systemd_service stop "${service%%-clone}" + check_resource_systemd "${service%%-clone}" started 600 +done diff --git a/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh b/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh index 7ed7012d..b8c5321b 100644 --- a/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh +++ b/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh @@ -56,3 +56,116 @@ function is_mysql_upgrade_needed { fi echo "1" } + +# This function returns the list of services to be migrated away from pacemaker +# and to systemd. The reason to have these services in a separate function is because +# this list is needed in three different places: major_upgrade_controller_pacemaker_{1,2} +# and in the function to migrate the cluster from full HA to HA NG +function services_to_migrate { + # The following PCMK resources the ones the we are going to delete + PCMK_RESOURCE_TODELETE=" + httpd-clone + memcached-clone + mongod-clone + neutron-dhcp-agent-clone + neutron-l3-agent-clone + neutron-metadata-agent-clone + neutron-netns-cleanup-clone + neutron-openvswitch-agent-clone + neutron-ovs-cleanup-clone + neutron-server-clone + openstack-aodh-evaluator-clone + openstack-aodh-listener-clone + openstack-aodh-notifier-clone + openstack-ceilometer-api-clone + openstack-ceilometer-central-clone + openstack-ceilometer-collector-clone + openstack-ceilometer-notification-clone + openstack-cinder-api-clone + openstack-cinder-scheduler-clone + openstack-glance-api-clone + openstack-glance-registry-clone + openstack-gnocchi-metricd-clone + openstack-gnocchi-statsd-clone + openstack-heat-api-cfn-clone + openstack-heat-api-clone + openstack-heat-api-cloudwatch-clone + openstack-heat-engine-clone + openstack-nova-api-clone + openstack-nova-conductor-clone + openstack-nova-consoleauth-clone + openstack-nova-novncproxy-clone + openstack-nova-scheduler-clone + openstack-sahara-api-clone + openstack-sahara-engine-clone + " + echo $PCMK_RESOURCE_TODELETE +} + +# This function will migrate a mitaka system where all the resources are managed +# via pacemaker to a newton setup where only a few services will be managed by pacemaker +# On a high-level it will operate as follows: +# 1. Set the cluster in maintenance-mode so no start/stop action will actually take place +# during the conversion +# 2. Remove all the colocation constraints and then the ordering constraints, except the +# ones related to haproxy/VIPs which exist in Newton as well +# 3. Remove all the resources that won't be managed by pacemaker in newton. Note that they +# will show up as ORPHANED but they will keep running normally via systemd. They will be +# enabled to start at boot by puppet during the converge step +# 4. Take the cluster out of maintenance-mode and do a resource cleanup +function migrate_full_to_ng_ha { + if [[ -n $(pcmk_running) ]]; then + pcs property set maintenance-mode=true + # We are making sure here that the property has propagated everywhere + if ! timeout -k 10 300 crm_resource --wait; then + echo_error "ERROR: cluster remained unstable after setting maintenance-mode for more than 300 seconds, exiting." + exit 1 + fi + # First we go through all the colocation constraints (except the ones we want to keep, i.e. the haproxy/ip ones) + # and we remove those + COL_CONSTRAINTS=$(pcs config show | sed -n '/^Colocation Constraints:$/,/^$/p' | grep -v "Colocation Constraints:" | egrep -v "ip-.*haproxy" | awk '{print $NF}' | cut -f2 -d: |cut -f1 -d\)) + for constraint in $COL_CONSTRAINTS; do + log_debug "Deleting colocation constraint $constraint from CIB" + pcs constraint remove "$constraint" + done + + # Now we kill all the ordering constraints (except the haproxy/ip ones) + ORD_CONSTRAINTS=$(pcs config show | sed -n '/^Ordering Constraints:/,/^Colocation Constraints:$/p' | grep -v "Ordering Constraints:" | awk '{print $NF}' | cut -f2 -d: |cut -f1 -d\)) + for constraint in $ORD_CONSTRAINTS; do + log_debug "Deleting ordering constraint $constraint from CIB" + pcs constraint remove "$constraint" + done + + # At this stage there are no constraints whatsoever except the haproxy/ip ones + # which we want to keep. We now delete each resource that will move to systemd + # Note that the corresponding systemd resource will stay running, which means that + # later when we do the "yum update", things will be a bit slower because each + # "systemctl try-restart " is not a no-op any longer because the service is up + # and running and it will be restarted with rabbitmq being down. + PCS_STATUS_OUTPUT="$(pcs status)" + for resource in $(services_to_migrate) "delay-clone" "openstack-core-clone"; do + if echo "$PCS_STATUS_OUTPUT" | grep "$resource"; then + log_debug "Deleting $resource from the CIB" + + # We need to add --force because the cluster is in maintenance mode and the resource + # is unmanaged. The if serves to make this idempotent + pcs resource delete --force "$resource" + else + log_debug "Service $service not found as a pacemaker resource, not trying to delete." + fi + done + + # At this stage all the pacemaker resources are removed from the CIB. Once we remove the + # maintenance-mode those systemd resources will keep on running. They shall be systemd enabled + # via the puppet converge step later on + pcs property set maintenance-mode=false + # We need to do a pcs resource cleanup here + crm_resource --wait to make sure the + # cluster is in a clean state before we stop everything, upgrade and restart everything + pcs resource cleanup + # We are making sure here that the cluster is stable before proceeding + if ! timeout -k 10 600 crm_resource --wait; then + echo_error "ERROR: cluster remained unstable after resource cleanup for more than 600 seconds, exiting." + exit 1 + fi + fi +} -- cgit 1.2.3-korg