Merge pull request #808 from hmcts/DTSPO-19088

DTSPO-19088 - Adding new AKS status script

.github/workflows/aks-auto-shutdown.yaml

@@ -65,4 +65,4 @@ jobs:
 
       - name: AKS Auto Shutdown status check
         if: env.DEV_ENV != 'true'
-        run: ./scripts/aks/auto-shutdown-status.sh stop ${{ secrets.REGISTRYSLACKWEBHOOK }}
+        run: ./scripts/aks/auto-shutdown-status.sh stop ${{ secrets.SHUTDOWN_NOTIFICATIONS_WEBHOOK }}

.github/workflows/aks-auto-start.yaml

@@ -65,4 +65,4 @@ jobs:
 
       - name: AKS Auto Shutdown status check
         #if: env.DEV_ENV != 'true'
-        run: ./scripts/aks/auto-shutdown-status.sh start ${{ secrets.REGISTRYSLACKWEBHOOK }}
+        run: ./scripts/aks/auto-shutdown-status.sh start ${{ secrets.SHUTDOWN_NOTIFICATIONS_WEBHOOK }}

scripts/aks/auto-shutdown-status.sh

@@ -1,29 +1,78 @@
 #!/usr/bin/env bash
-#set -x
+# set -x
 shopt -s nocasematch
-AMBER='\033[1;33m'
-GREEN='\033[0;32m'
 
+# Source shared function scripts
 source scripts/aks/common-functions.sh
 source scripts/common/common-functions.sh
 
+# Set variables for later use, MODE has a default but can be overridden at usage time
+# notificationSlackWebhook is used during the function call `auto_shutdown_notification`
 MODE=${1:-start}
-registrySlackWebhook=$2
+notificationSlackWebhook=$2
+SKIP="false"
+
+# Catch problems with MODE input, must be one of Start/Stop
+if [[ "$MODE" != "start" && "$MODE" != "stop" ]]; then
+    echo "Invalid MODE. Please use 'start' or 'stop'."
+    exit 1
+fi
 
 CLUSTERS=$(get_clusters)
 clusters_count=$(jq -c -r '.count' <<<$CLUSTERS)
-log "$clusters_count AKS Clusters found"
-log "----------------------------------------------"
+ts_echo "$clusters_count AKS Clusters found"
 
+# For each AKS Cluster found in the function `get_clusters` start another loop
 jq -c '.data[]' <<<$CLUSTERS | while read cluster; do
+    # Function that returns the Resource Group, Id and Name of the AKS Cluster and its current state as variables
     get_cluster_details
 
-    if [[ $cluster_status == "Stopped" ]]; then
-        echo -e "${GREEN}$CLUSTER_NAME is $cluster_status"
-    elif [[ $cluster_status == "Running" ]]; then
-        echo -e "${AMBER}$CLUSTER_NAME is $cluster_status"
+    # Set variables based on inputs which are used to decide when to SKIP an environment
+    if [[ $cluster_env == "sbox" ]]; then
+        cluster_env=${cluster_env/#sbox/Sandbox}
+    elif [[ $cluster_env == "ptlsbox" ]]; then
+        cluster_env=${cluster_env/ptlsbox/Sandbox}
+    elif [[ $cluster_env == "stg" ]]; then
+        cluster_env=${cluster_env/stg/Staging}
     fi
-    if [[ $MODE == "start" ]]; then
-        check_cluster_status
+
+    cluster_business_area=$(echo $CLUSTER_NAME | cut -d'-' -f1)
+    cluster_business_area=${cluster_business_area/ss/cross-cutting}
+
+    # SKIP variable updated based on the output of the `should_skip_start_stop` function which calculates its value
+    # based on the issues_list.json file which contains user requests to keep environments online after normal hours
+    SKIP=$(should_skip_start_stop $cluster_env $cluster_business_area $MODE)
+
+    # Setup message output templates for later use
+    logMessage="Cluster: $CLUSTER_NAME in Subscription: $SUBSCRIPTION  ResourceGroup: $RESOURCE_GROUP is in $CLUSTER_STATUS state after $MODE action"
+    slackMessage="Cluster: *$CLUSTER_NAME* in Subscription: *$SUBSCRIPTION* is in *$CLUSTER_STATUS* state after *$MODE* action"
+
+    # If SKIP is false then we progress with the status check for the particular Flexible server in this loop run, if SKIP is true then do nothing
+    if [[ $SKIP == "false" ]]; then
+        # Check state of the AKS Cluster and print output as required
+        # Depending on the value of MODE a notification will also be sent
+        #    - If MODE = Start then a stopped AKS Cluster is incorrect and we should notify
+        #    - If MODE = Stop then a running AKS Cluster is incorrect and we should notify
+        #    - If neither Running or Stopped is found then something else is going on and we should notify
+        case "$CLUSTER_STATUS" in
+        *"Running"*)
+            ts_echo_color $([[ $MODE == "start" ]] && echo GREEN || echo RED) "$logMessage"
+            if [[ $MODE == "stop" ]]; then
+                auto_shutdown_notification ":red_circle: $slackMessage"
+            fi
+            ;;
+        *"Stopped"*)
+            ts_echo_color $([[ $MODE == "start" ]] && echo RED || echo GREEN) "$logMessage"
+            if [[ $MODE == "start" ]]; then
+                auto_shutdown_notification ":red_circle: $slackMessage"
+            fi
+            ;;
+        *)
+            ts_echo_color AMBER "$logMessage"
+            auto_shutdown_notification ":yellow_circle: $slackMessage"
+            ;;
+        esac
+    else
+        ts_echo_color AMBER "Cluster: $SERVER_NAME in ResourceGroup: $RESOURCE_GROUP has been skipped from today's $MODE operation schedule"
     fi
-done
+done

scripts/aks/common-functions.sh

@@ -27,7 +27,7 @@ function get_cluster_details() {
     RESOURCE_GROUP=$(jq -r '.resourceGroup' <<<$cluster)
     CLUSTER_NAME=$(jq -r '.name' <<<$cluster)
     STARTUP_MODE=$(jq -r '.tags.startupMode' <<<$cluster)
-    CLUSTER_STATUS=$(jq -r '.powerState.code' <<<$cluster)
+    CLUSTER_STATUS=$(jq -r '.properties.powerState.code' <<<$cluster)
     SUBSCRIPTION=$(jq -r '.subscriptionId' <<<$cluster)
 }