Merge "Select per-network hostnames for service_node_names"

[apex-tripleo-heat-templates.git] / extraconfig / tasks / major_upgrade_pacemaker_migrations.sh

diff --git a/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh b/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh
index d974bb7..df87c93 100644 (file)
--- a/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh
+++ b/extraconfig/tasks/major_upgrade_pacemaker_migrations.sh
@@ -109,7 +109,7 @@ function services_to_migrate {
 #    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 and do a resource cleanup
+# 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
@@ -117,13 +117,9 @@ function services_to_migrate {
 function migrate_full_to_ng_ha {
     if [[ -n $(pcmk_running) ]]; then
         pcs property set maintenance-mode=true
-        # We are making sure here that the property has propagated everywhere
-        if ! timeout -k 10 300 crm_resource --wait; then
-            echo_error "ERROR: cluster remained unstable after setting maintenance-mode for more than 300 seconds, exiting."
-            exit 1
-        fi
-        # First we go through all the colocation constraints (except the ones we want to keep, i.e. the haproxy/ip ones)
-        # and we remove those
+
+        # First we go through all the colocation constraints (except the ones
+        # we want to keep, i.e. the haproxy/ip ones) and we remove those
         COL_CONSTRAINTS=$(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 $constraint from CIB"
@@ -158,7 +154,7 @@ function migrate_full_to_ng_ha {
                  fi
                  pcs resource delete --force "$resource"
              else
-                 log_debug "Service $service not found as a pacemaker resource, not trying to delete."
+                 log_debug "Service $resource not found as a pacemaker resource, not trying to delete."
              fi
         done
 
@@ -173,3 +169,22 @@ function migrate_full_to_ng_ha {
         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
+}