X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?a=blobdiff_plain;f=jjb%2Fopendaylight-infra-stack.sh;h=3a105144081b831bc6587ef23aa1de7e75a804e9;hb=f2a032fec39a544dae44694960b99bb0434e9020;hp=4d4ff6e2dea7c1854896486d289379d27baca8a5;hpb=7885c1966559e25710ecbd0ab477e96514c9b208;p=releng%2Fbuilder.git

diff --git a/jjb/opendaylight-infra-stack.sh b/jjb/opendaylight-infra-stack.sh
index 4d4ff6e2d..3a1051440 100644
--- a/jjb/opendaylight-infra-stack.sh
+++ b/jjb/opendaylight-infra-stack.sh
@@ -1,19 +1,89 @@
 #!/bin/bash
-git clone --depth 1 https://git.opendaylight.org/gerrit/releng/builder $WORKSPACE/.infra-scripts
-
-virtualenv $WORKSPACE/.venv-openstack
-source $WORKSPACE/.venv-openstack/bin/activate
+virtualenv "$WORKSPACE/.venv-openstack"
+# shellcheck disable=SC1090
+source "$WORKSPACE/.venv-openstack/bin/activate"
 pip install --upgrade pip
 pip install --upgrade python-openstackclient python-heatclient
 pip freeze
 
-cd $WORKSPACE/.infra-scripts/openstack-hot
+cd /builder/openstack-hot || exit 1
 
-JOB_SUM=`echo $JOB_NAME | sum | awk '{{ print $1 }}'`
+JOB_SUM=$(echo "$JOB_NAME" | sum | awk '{{ print $1 }}')
 VM_NAME="$JOB_SUM-$BUILD_NUMBER"
-openstack --os-cloud rackspace stack create --wait -t {stack-template} -e $WORKSPACE/opendaylight-infra-environment.yaml --parameter "job_name=$VM_NAME" --parameter "silo=$SILO" $STACK_NAME
-OS_STATUS=`openstack --os-cloud rackspace stack show -f json -c stack_status $STACK_NAME | jq -r '.stack_status'`
-if [ "$OS_STATUS" != "CREATE_COMPLETE" ]; then
-    echo "Failed to initialize infrastructure. Quitting..."
+
+OS_TIMEOUT=10  # Minutes to wait for OpenStack VM to come online
+STACK_RETRIES=3  # Number of times to retry creating a stack before fully giving up
+STACK_SUCCESSFUL=false
+# seq X refers to waiting for X minutes for OpenStack to return
+# a status that is not CREATE_IN_PROGRESS before giving up.
+openstack --os-cloud rackspace limits show --absolute
+openstack --os-cloud rackspace limits show --rate
+echo "Trying up to $STACK_RETRIES times to create $STACK_NAME."
+for try in $(seq $STACK_RETRIES); do
+    # shellcheck disable=SC1083
+    openstack --os-cloud rackspace stack create --timeout "$OS_TIMEOUT" -t {stack-template} -e "$WORKSPACE/opendaylight-infra-environment.yaml" --parameter "job_name=$VM_NAME" --parameter "silo=$SILO" "$STACK_NAME"
+    openstack --os-cloud rackspace stack list
+    echo "$try: Waiting for $OS_TIMEOUT minutes to create $STACK_NAME."
+    for i in $(seq $OS_TIMEOUT); do
+        sleep 60
+        OS_STATUS=$(openstack --os-cloud rackspace stack show -f json -c stack_status "$STACK_NAME" | jq -r '.stack_status')
+        echo "$i: $OS_STATUS"
+
+        case "$OS_STATUS" in
+            CREATE_COMPLETE)
+                echo "Stack initialized on infrastructure successful."
+                STACK_SUCCESSFUL=true
+                break
+            ;;
+            CREATE_FAILED)
+                echo "ERROR: Failed to initialize infrastructure. Deleting stack and possibly retrying to create..."
+                openstack --os-cloud rackspace stack list
+                openstack --os-cloud rackspace stack delete --yes "$STACK_NAME"
+                openstack --os-cloud rackspace stack show "$STACK_NAME"
+                # after stack delete, poll for 10m to know when stack is fully removed
+                # the logic here is that when "stack show $STACK_NAME" does not contain $STACK_NAME
+                # we assume it's successfully deleted and we can break to retry
+                for j in $(seq 20); do
+                    sleep 30;
+                    STACK_SHOW=$(openstack --os-cloud rackspace stack show "$STACK_NAME")
+                    echo "$j: $STACK_SHOW"
+                    if [[ $STACK_SHOW == *"DELETE_FAILED"* ]]; then
+                        echo "stack delete failed. trying to stack abandon now"
+                        openstack --os-cloud rackspace stack abandon "$STACK_NAME"
+                        STACK_SHOW=$(openstack --os-cloud rackspace stack show "$STACK_NAME")
+                        echo "$STACK_SHOW"
+                    fi
+                    if [[ $STACK_SHOW != *"$STACK_NAME"* ]]; then
+                        echo "stack show on $STACK_NAME came back empty. Assuming successful delete"
+                        break
+                    fi
+                done
+                # if we still see $STACK_NAME in $STACK_SHOW it means the delete hasn't fully
+                # worked and we can exit forcefully
+                if [[ $STACK_SHOW == *"$STACK_NAME"* ]]; then
+                    echo "stack $STACK_NAME still in stack show output after polling. Quitting!"
+                    exit 1
+                fi
+                break
+            ;;
+            CREATE_IN_PROGRESS)
+                echo "Waiting to initialize infrastructure."
+                continue
+            ;;
+            *)
+                echo "Unexpected status: $OS_STATUS"
+                exit 1
+            ;;
+        esac
+    done
+    if $STACK_SUCCESSFUL; then
+        break
+    fi
+done
+
+# capture stack info in console logs
+openstack --os-cloud rackspace stack show "$STACK_NAME"
+
+if ! $STACK_SUCCESSFUL; then
     exit 1
 fi