X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?a=blobdiff_plain;f=jjb%2Fopendaylight-infra-stack.sh;h=dd731f1ac4b034609f0b153ce0a7d9f213f092d9;hb=29b2f25826a1d889aa515be1c39c40da83f7ae00;hp=a5418f5c934ea57d82dba8947a5cd41c6cdc9efb;hpb=b095da7fce0901d5b35a625ce3a0dab4d256be73;p=releng%2Fbuilder.git

diff --git a/jjb/opendaylight-infra-stack.sh b/jjb/opendaylight-infra-stack.sh
index a5418f5c9..dd731f1ac 100644
--- a/jjb/opendaylight-infra-stack.sh
+++ b/jjb/opendaylight-infra-stack.sh
@@ -1,17 +1,98 @@
 #!/bin/bash
-virtualenv $WORKSPACE/.venv-openstack
-source $WORKSPACE/.venv-openstack/bin/activate
+virtualenv "/tmp/v/openstack"
+# shellcheck source=/tmp/v/openstack/bin/activate disable=SC1091
+source "/tmp/v/openstack/bin/activate"
 pip install --upgrade pip
 pip install --upgrade python-openstackclient python-heatclient
 pip freeze
 
-cd /builder/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 limits show --absolute
+openstack limits show --rate
+echo "Trying up to $STACK_RETRIES times to create $STACK_NAME."
+for try in $(seq $STACK_RETRIES); do
+    # shellcheck disable=SC1083
+    openstack stack create --timeout "$OS_TIMEOUT" -t {stack-template} -e "$WORKSPACE/opendaylight-infra-environment.yaml" --parameter "job_name=$VM_NAME" --parameter "silo=$SILO" "$STACK_NAME"
+    echo "$try: Waiting for $OS_TIMEOUT minutes to create $STACK_NAME."
+    for i in $(seq $OS_TIMEOUT); do
+        sleep 60
+        OS_STATUS=$(openstack 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 stack delete --yes "$STACK_NAME"
+                openstack stack show "$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 stack show "$STACK_NAME")
+                    echo "$j: $STACK_SHOW"
+                    if [[ $STACK_SHOW == *"DELETE_FAILED"* ]]; then
+                        echo "stack delete failed. trying to stack abandon now"
+                        # stack abandon does not work on RS, therefore requires acquiring a token
+                        # and using http delete method to abondon DELETE_FAILED stacks
+                        # Todo: remove the change once RS fixes the issue upstream
+                        # openstack stack abandon "$STACK_NAME"
+                        STACK_ID=$(openstack stack show -f json -c "id" "$STACK_NAME" | jq -r '."id"')
+                        TOKEN=$(openstack token issue -f json -c id | jq -r '.id')
+                        curl -si -X DELETE -H "Content-Type: application/json" -H "Accept: application/json"\
+                            -H "x-auth-token: $TOKEN"\
+                            "https://dfw.orchestration.api.rackspacecloud.com/v1/904885/stacks/$STACK_NAME/$STACK_ID/abandon"
+                        STACK_SHOW=$(openstack 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"
+                # DO NOT exit on unexpected status. Rackspace sometimes returns unexpected status
+                # before returning an expected status. Just print the message and loop until we have
+                # a confirmed state or timeout.
+                # exit 1
+            ;;
+        esac
+    done
+    if $STACK_SUCCESSFUL; then
+        break
+    fi
+done
+
+# capture stack info in console logs
+openstack stack show "$STACK_NAME"
+
+if ! $STACK_SUCCESSFUL; then
     exit 1
 fi