Update prometheus-grafana extension to run on agent nodes (#1959)

* Update prometheus-grafana extension to run on agent nodes

* Support running on both master and agent nodes

* Update example json making run on agent default

* add retry logic to initial helm install for prometheus chart

* Update rootURL to point to fork for CI

* Revert to upsteam for rootURL

examples/extensions/prometheus-grafana-k8s.json

@@ -7,19 +7,19 @@
     "masterProfile": {
       "count": 1,
       "dnsPrefix": "",
-      "vmSize": "Standard_DS2_v2",
-      "extensions": [
-        { 
-          "name": "prometheus-grafana-k8s"
-        }
-      ]
+      "vmSize": "Standard_DS2_v2"
     },
     "agentPoolProfiles": [
       {
         "name": "agentpool1",
         "count": 3,
         "vmSize": "Standard_DS2_v2",
-        "availabilityProfile": "AvailabilitySet"
+        "availabilityProfile": "AvailabilitySet",
+        "extensions": [
+          { 
+            "name": "prometheus-grafana-k8s"
+          }
+        ]
       }
     ],
     "linuxProfile": {

extensions/prometheus-grafana-k8s/v1/prometheus-grafana-k8s.sh

@@ -3,6 +3,9 @@ set -x
 
 echo $(date) " - Starting Script"
 
+echo $(date) " - Setting kubeconfig"
+export KUBECONFIG=/var/lib/kubelet/kubeconfig
+
 echo $(date) " - Waiting for API Server to start"
 kubernetesStarted=1
 for i in {1..600}; do
@@ -59,15 +62,48 @@ wait_for_master_nodes() {
     return 1
 }
 
+agent_nodes() {
+    kubectl get no -L kubernetes.io/role -l kubernetes.io/role=agent --no-headers -o jsonpath="{.items[*].metadata.name}" | tr " " "\n" | sort | head -n 1
+}
+
+wait_for_agent_nodes() {
+    ATTEMPTS=90
+    SLEEP_TIME=10
+
+    ITERATION=0
+    while [[ $ITERATION -lt $ATTEMPTS ]]; do
+        echo $(date) " - Is kubectl returning agent nodes? (attempt $(( $ITERATION + 1 )) of $ATTEMPTS)"
+
+        FIRST_K8S_AGENT=$(agent_nodes)
+
+        if [[ -n $FIRST_K8S_AGENT ]]; then
+            echo $(date) " - kubectl is returning agent nodes"
+            return
+        fi
+
+        ITERATION=$(( $ITERATION + 1 ))
+        sleep $SLEEP_TIME
+    done
+
+    echo $(date) " - kubectl failed to return agent nodes in the alotted time"
+    return 1
+}
+
 should_this_node_run_extension() {
     FIRST_K8S_MASTER=$(master_nodes)
     if [[ $FIRST_K8S_MASTER = $(hostname) ]]; then
         echo $(date) " - Local node $(hostname) is found to be the first master node $FIRST_K8S_MASTER"
         return
     else
-        echo $(date) " - Local node $(hostname) is not the first master node $FIRST_K8S_MASTER"
+        FIRST_K8S_AGENT=$(agent_nodes)
+        if [[ $FIRST_K8S_AGENT = $(hostname) ]]; then
+            echo $(date) " - Local node $(hostname) is found to be the first agent node $FIRST_K8S_AGENT"
+            return
+        else
+            echo $(date) " - Local node $(hostname) is not the first master node $FIRST_K8S_MASTER or the first agent node $FIRST_K8S_AGENT"
             return 1
         fi
+    fi
 }
 
 storageclass_param() {
@@ -129,16 +165,29 @@ install_prometheus() {
 
     echo $(date) " - Installing the Prometheus Helm chart"
 
+    ATTEMPTS=90
+    SLEEP_TIME=10
+
+    ITERATION=0
+    while [[ $ITERATION -lt $ATTEMPTS ]]; do
         helm install -f prometheus_values.yaml \
             --name $PROM_RELEASE_NAME \
             --namespace $NAMESPACE stable/prometheus $(storageclass_param)
 
+        if [[ $? -eq 0 ]]; then
+            echo $(date) " - Helm install successfully completed"
+            break
+        else
+            echo $(date) " - Helm install returned a non-zero exit code. Retrying."
+        fi
+
+        ITERATION=$(( $ITERATION + 1 ))
+        sleep $SLEEP_TIME
+    done
+
     PROM_POD_PREFIX="$PROM_RELEASE_NAME-prometheus-server"
     DESIRED_POD_STATE=Running
 
-    ATTEMPTS=90
-    SLEEP_TIME=10
-
     ITERATION=0
     while [[ $ITERATION -lt $ATTEMPTS ]]; do
         echo $(date) " - Is the prometheus server pod ($PROM_POD_PREFIX-*) running? (attempt $(( $ITERATION + 1 )) of $ATTEMPTS)"
@@ -212,9 +261,16 @@ if [[ $? -ne 0 ]]; then
     echo $(date) " - Error while waiting for kubectl to output master nodes. Exiting"
     exit 1
 fi
+
+wait_for_agent_nodes
+if [[ $? -ne 0 ]]; then
+    echo $(date) " - Error while waiting for kubectl to output agent nodes. Exiting"
+    exit 1
+fi
+
 should_this_node_run_extension
 if [[ $? -ne 0 ]]; then
-    echo $(date) " - Not the first master node, no longer continuing extension. Exiting"
+    echo $(date) " - Not the first master node or the first agent node, no longer continuing extension. Exiting"
     exit 1
 fi