Changed cpu and memory metric stat to avg (#173)

Signed-off-by: Rishabh Singh <[email protected]>
opensearch-project · Jul 27, 2022 · 72273fa · 72273fa
1 parent 4da4009
commit 72273fa
Show file tree

Hide file tree

Showing 3 changed files with 34 additions and 3 deletions.
diff --git a/lib/compute/jenkins-main-node.ts b/lib/compute/jenkins-main-node.ts
@@ -337,7 +337,7 @@ export class JenkinsMainNode {
           metrics_collected: {
             procstat: [
               {
-                pattern: 'jenkins',
+                pattern: 'Djenkins',
                 measurement: [
                   'cpu_usage',
                   'cpu_time_system',

diff --git a/lib/monitoring/ci-alarms.ts b/lib/monitoring/ci-alarms.ts
@@ -39,7 +39,7 @@ export class JenkinsMonitoring {
 
     this.alarms.push(new Alarm(stack, 'MainNodeHighCpuUtilization', {
       alarmDescription: 'The jenkins process is using much more CPU that expected, it should be investigated for a stuck process/job',
-      metric: mainNode.ec2InstanceMetrics.cpuTime.with({ statistic: 'max' }),
+      metric: mainNode.ec2InstanceMetrics.cpuTime.with({ statistic: 'avg' }),
       evaluationPeriods: 5,
       threshold: 50,
       comparisonOperator: ComparisonOperator.GREATER_THAN_OR_EQUAL_TO_THRESHOLD,
@@ -48,7 +48,7 @@ export class JenkinsMonitoring {
 
     this.alarms.push(new Alarm(stack, 'MainNodeHighMemoryUtilization', {
       alarmDescription: 'The jenkins process is using more memory than expected, it should be investigated for a large number of jobs or heavy weight jobs',
-      metric: mainNode.ec2InstanceMetrics.memUsed.with({ statistic: 'max' }),
+      metric: mainNode.ec2InstanceMetrics.memUsed.with({ statistic: 'avg' }),
       evaluationPeriods: 5,
       threshold: 50,
       comparisonOperator: ComparisonOperator.GREATER_THAN_OR_EQUAL_TO_THRESHOLD,

diff --git a/test/ci-stack.test.ts b/test/ci-stack.test.ts
@@ -35,6 +35,7 @@ test('CI Stack Basic Resources', () => {
   expect(stack).to(countResources('AWS::SSM::Document', 1));
   expect(stack).to(countResources('AWS::SSM::Association', 1));
   expect(stack).to(countResources('AWS::EFS::FileSystem', 1));
+  expect(stack).to(countResources('AWS::CloudWatch::Alarm', 5));
 });
 
 test('External security group is open', () => {
@@ -141,3 +142,33 @@ test('LoadBalancer', () => {
     ],
   }, ResourcePart.Properties));
 });
+
+test('CloudwatchCpuAlarm', () => {
+  const app = new App({
+    context: { useSsl: 'false', runWithOidc: 'false' },
+  });
+
+  // WHEN
+  const stack = new CIStack(app, 'MyTestStack', {});
+
+  // THEN
+  expect(stack).to(haveResourceLike('AWS::CloudWatch::Alarm', {
+    MetricName: 'procstat_cpu_usage',
+    Statistic: 'Average',
+  }, ResourcePart.Properties));
+});
+
+test('CloudwatchMemoryAlarm', () => {
+  const app = new App({
+    context: { useSsl: 'false', runWithOidc: 'false' },
+  });
+
+  // WHEN
+  const stack = new CIStack(app, 'MyTestStack', {});
+
+  // THEN
+  expect(stack).to(haveResourceLike('AWS::CloudWatch::Alarm', {
+    MetricName: 'mem_used_percent',
+    Statistic: 'Average',
+  }, ResourcePart.Properties));
+});