Check for stacks delete in progress

[releng/builder.git] / jjb / opendaylight-infra-cleanup-stale-stacks.sh

diff --git a/jjb/opendaylight-infra-cleanup-stale-stacks.sh b/jjb/opendaylight-infra-cleanup-stale-stacks.sh
index f44ac51079794a844d02635f17ade872e8ec106f..d99938b3444dbbc09a741e9dc4305eba0c4334f9 100644 (file)
--- a/jjb/opendaylight-infra-cleanup-stale-stacks.sh
+++ b/jjb/opendaylight-infra-cleanup-stale-stacks.sh
@@ -1,16 +1,74 @@
 #!/bin/bash
-virtualenv $WORKSPACE/.venv
-source $WORKSPACE/.venv/bin/activate
+virtualenv "$WORKSPACE/.venv"
+# shellcheck disable=SC1090
+source "$WORKSPACE/.venv/bin/activate"
 pip install --upgrade pip
 pip install --upgrade python-openstackclient python-heatclient
 pip freeze
 
-DELETE_LIST=(`openstack --os-cloud rackspace stack list -f json | \
-              jq -r '.[] | \
-                     select((."Stack Status" == "CREATE_FAILED") or \
-                            (."Stack Status" == "DELETE_FAILED")) | \
-                     ."Stack Name"'`)
-for i in "${DELETE_LIST[@]}"; do
-    echo "Deleting stack $i"
-    openstack --os-cloud rackspace stack delete --yes $i
+#########################
+## FETCH ACTIVE BUILDS ##
+#########################
+# Fetch stack list before fetching active builds to minimize race condition
+# where we might be try to delete stacks while jobs are trying to start
+OS_STACKS=($(openstack stack list \
+            -f json -c "Stack Name" -c "Stack Status" \
+            --property "stack_status=CREATE_COMPLETE" \
+            --property "stack_status=DELETE_FAILED" \
+            --property "stack_status=CREATE_FAILED" \
+            | jq -r '.[] | ."Stack Name"'))
+
+# Make sure we fetch active builds on both the releng and sandbox silos
+ACTIVE_BUILDS=()
+for silo in releng sandbox; do
+    JENKINS_URL="https://jenkins.opendaylight.org/$silo//computer/api/json?tree=computer[executors[currentExecutable[url]],oneOffExecutors[currentExecutable[url]]]&xpath=//url&wrapper=builds"
+    wget --no-verbose -O "${silo}_builds.json" "$JENKINS_URL"
+    sleep 1  # Need to sleep for 1 second otherwise next line causes script to stall
+    ACTIVE_BUILDS=(${ACTIVE_BUILDS[@]} $( \
+        jq -r '.computer[].executors[].currentExecutable.url' "${silo}_builds.json" \
+        | grep -v null | awk -F'/' '{print $4 "-" $6 "-" $7}'))
+done
+
+##########################
+## DELETE UNUSED STACKS ##
+##########################
+# Search for stacks that are not in use by either releng or sandbox silos and
+# delete them.
+for STACK_NAME in "${OS_STACKS[@]}"; do
+    STACK_STATUS=$(openstack stack show -f json -c "stack_status" "$STACK_NAME" | jq -r '."stack_status"')
+    if [[ "${ACTIVE_BUILDS[@]}" =~ $STACK_NAME ]]; then
+        # No need to delete stacks if there exists an active build for them
+        continue
+    else
+        OS_STATUS=$(openstack stack show -f json -c stack_status "$stack" | jq -r '.stack_status')
+        case "$OS_STATUS" in
+            DELETE_IN_PROGRESS)
+                echo "skipping delete, $stack is already DELETE in progress."
+                continue
+            ;;
+            DELETE_FAILED)
+                echo "Stack delete failed, trying to stack abandon now."
+                # stack abandon does not work on RS, therefore requires acquiring a token
+                # and using http delete method to abondon DELETE_FAILED stacks
+                # Todo: remove the change once RS fixes the issue upstream
+                # openstack stack abandon "$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"
+                continue
+            ;;
+            CREATE_COMPLETE|CREATE_FAILED)
+                echo "Deleting orphaned stack: $stack"
+                openstack stack delete --yes "$stack"
+                continue
+            ;;
+            *)
+                continue
+            ;;
+        esac
+    fi
 done