Merge branch 'jeanschmidt/userdata_metrics' of https://github.com/pyt…

…orch/test-infra into jeanschmidt/userdata_metrics

terraform-aws-github-runner/modules/runners-instances/templates/user-data.sh

@@ -39,6 +39,43 @@ function err_report () {
 
 trap 'err_report $LINENO' ERR
 
+function metric_report () {
+    local metric_name=$1
+    local value=$2
+
+    # it is useful to not have a namespace and send all errors here, in case we can't get the environment for some reason
+    # this should not be the case, as this environment variable is set externally
+    # it is important to have the || true at the end, as we dont want to interrupt the scritp at failure and trigger an infinite loop
+    aws cloudwatch put-metric-data --metric-name "$metric_name" --namespace "GHARunners/all" --value $value || true
+
+    local namespace="GHARunners/all"
+    if [ ! -z "$environment"]; then
+        namespace="GHARunners/$environment"
+        aws cloudwatch put-metric-data --metric-name "$metric_name" --namespace "GHARunners/all" --value $value || true
+    fi
+
+    if [ ! -z "$INSTANCE_ID" ]; then
+        aws cloudwatch put-metric-data --metric-name "$metric_name" --namespace "$namespace" --value $value --dimensions "InstanceId=$INSTANCE_ID" || true
+    fi
+    if [ ! -z "$REGION" ]; then
+        aws cloudwatch put-metric-data --metric-name "$metric_name" --namespace "$namespace" --value $value --dimensions "Region=$REGION" || true
+    fi
+    if [ ! -z "$OS_ID" ]; then
+        aws cloudwatch put-metric-data --metric-name "$metric_name" --namespace "$namespace" --value $value --dimensions "os=$OS_ID" || true
+    fi
+    if [ ! -z "$OS_ID" ]; then
+        aws cloudwatch put-metric-data --metric-name "$metric_name" --namespace "$namespace" --value $value --dimensions GHRunnerId=$GH_RUNNER_ID || true
+    fi
+}
+
+function err_report () {
+    echo "Error on line $1"
+    metric_report "linux_userdata.error" 1
+    exit 1
+}
+
+trap 'err_report $LINENO' ERR
+
 function retry {
   local retries=7
   local count=0