1 files changed, 132 insertions, 0 deletions
diff --git a/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh b/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh
index 7ed7012d..df87c93f 100644
--- a/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh
+++ b/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh
@@ -56,3 +56,135 @@ 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. Take the cluster out of maintenance-mode
+# 4. Remove all the resources that won't be managed by pacemaker in newton. The
+#    outcome will be
+#    that they are stopped and removed from pacemakers control
+# 5. Do a resource cleanup to make sure the cluster is in a clean state
+function migrate_full_to_ng_ha {
+    if [[ -n $(pcmk_running) ]]; then
+        pcs property set maintenance-mode=true
+
+        # First we go through all the colocation constraints (except the ones
+        # we want to keep, i.e. the haproxy/ip ones) and we remove those
+        COL_CONSTRAINTS=$(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 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
+
+        # At this stage there are no constraints whatsoever except the haproxy/ip ones
+        # which we want to keep. We now disable and then delete each resource
+        # that will move to systemd.
+        # We want the systemd resources be stopped before doing "yum update",
+        # that way "systemctl try-restart <service>" is no-op because the
+        # service was down already 
+        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 $resource from the CIB"
+                 if ! pcs resource disable "$resource" --wait=600; then
+                     echo_error "ERROR: resource $resource failed to be disabled"
+                     exit 1
+                 fi
+                 pcs resource delete --force "$resource"
+             else
+                 log_debug "Service $resource not found as a pacemaker resource, not trying to delete."
+             fi
+        done
+
+        # We need to do a pcs resource cleanup here + crm_resource --wait to
+        # make sure the cluster is in a clean state before we stop everything,
+        # upgrade and restart everything
+        pcs resource cleanup
+        # We are making sure here that the cluster is stable before proceeding
+        if ! timeout -k 10 600 crm_resource --wait; then
+            echo_error "ERROR: cluster remained unstable after resource cleanup for more than 600 seconds, exiting."
+            exit 1
+        fi
+    fi
+}
+
+# This function will make sure that the rabbitmq ha policies are converted from mitaka to newton
+# In mitaka we had: Attributes: set_policy="ha-all ^(?!amq\.).* {"ha-mode":"all"}"
+# In newton we want: Attributes: set_policy="ha-all ^(?!amq\.).* {"ha-mode":"exactly","ha-params":2}"
+# The nr "2" should be CEIL(N/2) where N is the number of Controllers (i.e. rabbit instances)
+# Note that changing an attribute like this makes the rabbitmq resource restart
+function rabbitmq_mitaka_newton_upgrade {
+    if pcs resource show rabbitmq-clone | grep -q -E "Attributes:.*\"ha-mode\":\"all\""; then
+        # Number of controller is obtained by counting how many hostnames we
+        # have in controller_node_names hiera key
+        nr_controllers=$(($(hiera controller_node_names | grep -o "," |wc -l) + 1))
+        nr_queues=$(($nr_controllers / 2 + ($nr_controllers % 2)))
+        if ! [ $nr_queues -gt 0 -a $nr_queues -le $nr_controllers ]; then
+            echo_error "ERROR: The nr. of HA queues during the M/N upgrade is out of range $nr_queues"
+            exit 1
+        fi
+        pcs resource update rabbitmq set_policy='ha-all ^(?!amq\\.).* {"ha-mode":"exactly","ha-params":'"$nr_queues}" --wait=600
+    fi
+}