Fix tests in preparation for making otel self monitoring default (#10537) (#10553)

* Fix leak integration test

* Improve detection of agent processes in integration tests

* Explicitly set the runtime in monitoring integration test

* Make the check for component metrics more generic

* Check the right build hash in integration tests

(cherry picked from commit 3ccf28d)

Co-authored-by: Mikołaj Świątek <[email protected]>

Author: mergify[bot]

pkg/testing/fixture_install.go

@@ -409,8 +409,23 @@ func mapProcess(p agentsystemprocess.ProcState) runningProcess {
 	return mappedProcess
 }
 
+// getElasticAgentProcesses returns the running elastic-agent control plane processes. Note that this does not mean
+// all processes running using the elastic-agent binary.
 func getElasticAgentProcesses(t *gotesting.T) []runningProcess {
-	return getProcesses(t, `.*elastic\-agent.*`)
+	agentProcesses := getProcesses(t, `.*elastic\-agent.*`)
+	// the otel collector might be running as a subprocess using the elastic-agent otel command
+	// we don't want to count these, so we drop all processes which are children of another process on the list
+	pids := make(map[int]struct{}, len(agentProcesses))
+	for _, p := range agentProcesses {
+		pids[p.Pid] = struct{}{}
+	}
+	agentControlPlaneProcesses := make([]runningProcess, 0, len(agentProcesses))
+	for _, p := range agentProcesses {
+		if _, ok := pids[p.Ppid]; !ok {
+			agentControlPlaneProcesses = append(agentControlPlaneProcesses, p)
+		}
+	}
+	return agentControlPlaneProcesses
 }
 
 // Includes both the main elastic-agent process and the agentbeat sub-processes for ensuring

testing/integration/ess/beat_receivers_test.go

@@ -219,11 +219,12 @@ func TestClassicAndReceiverAgentMonitoring(t *testing.T) {
 	require.NoError(t, err, "error unmarshalling policy: %s", string(policyBytes))
 	d, prs := policy.Outputs["default"]
 	require.True(t, prs, "default must be in outputs")
-	d.ApiKey = string(apiKey)
+	d.ApiKey = apiKey
 	policy.Outputs["default"] = d
 
 	processNamespace := fmt.Sprintf("%s-%s", info.Namespace, "process")
 	policy.Agent.Monitoring["namespace"] = processNamespace
+	policy.Agent.Monitoring["_runtime_experimental"] = "process"
 
 	updatedPolicyBytes, err := yaml.Marshal(policy)
 	require.NoErrorf(t, err, "error marshalling policy, struct was %v", policy)

testing/integration/ess/metrics_monitoring_test.go

@@ -172,8 +172,7 @@ func (runner *MetricsRunner) TestBeatsMetrics() {
 			query = genESQuery(agentStatus.Info.ID,
 				[][]string{
 					{"match", "component.id", cid},
-					{"exists", "field", "system.process.cpu.total.value"},
-					{"exists", "field", "system.process.memory.size"},
+					{"match", "agent.type", "metricbeat"},
 				})
 			now = time.Now()
 			res, err := estools.PerformQueryForRawQuery(ctx, query, "metrics-elastic_agent*", runner.info.ESClient)

testing/integration/ess/package_version_test.go

@@ -194,7 +194,12 @@ func TestComponentBuildHashInDiagnostics(t *testing.T) {
 	require.Lenf(t, splits, 2,
 		"expected split of beats output version to be split into 2, it was split into %q",
 		strings.Join(splits, "|"))
-	wantBuildHash := splits[0]
+	filebeatBuildHash := splits[0]
+
+	// get the elastic-agent build hash
+	agentVersion, err := f.ExecVersion(ctx)
+	require.NoError(t, err, "failed to get agent version")
+	agentCommit := agentVersion.Binary.Commit
 
 	diagZip, err := f.ExecDiagnostics(ctx)
 	require.NoError(t, err, "failed collecting diagnostics")
@@ -251,9 +256,20 @@ func TestComponentBuildHashInDiagnostics(t *testing.T) {
 	require.NoError(t, err, "could not parse state.yaml (%s)", stateYAML.Name())
 
 	for _, c := range state.Components {
-		assert.Equalf(t, wantBuildHash, c.State.VersionInfo.BuildHash,
+		var expectedCommit, expectedBuildHash string
+		switch c.State.VersionInfo.Name {
+		case "beats-receiver":
+			expectedBuildHash = agentCommit
+			expectedCommit = agentCommit
+		case "beat-v2-client":
+			expectedBuildHash = filebeatBuildHash
+			expectedCommit = filebeatBuildHash
+		default:
+			t.Errorf("got unknown value in version_info.name: %s", c.State.VersionInfo.Name)
+		}
+		assert.Equalf(t, expectedBuildHash, c.State.VersionInfo.BuildHash,
 			"component %s: VersionInfo.BuildHash mismatch", c.ID)
-		assert.Equalf(t, wantBuildHash, c.State.VersionInfo.Meta.Commit,
+		assert.Equalf(t, expectedCommit, c.State.VersionInfo.Meta.Commit,
 			"component %s: VersionInfo.Meta.Commit mismatch", c.ID)
 	}
 

testing/integration/leak/long_running_test.go

@@ -382,6 +382,9 @@ func (handleMon *handleMonitor) Init(ctx context.Context, t *testing.T, fixture
 
 	for _, comp := range status.Components {
 		pidStr := pidInStatusMessageRegex.FindString(comp.Message)
+		if pidStr == "" { // could be an otel receiver, not a process
+			continue
+		}
 		pid, err := strconv.ParseInt(pidStr, 10, 64)
 		require.NoError(t, err)