Fix a few issues with testing, add better cluster information

Author: HoustonPutman

controllers/solr_cluster_ops_util.go

@@ -66,6 +66,7 @@ func clearClusterOpLock(statefulSet *appsv1.StatefulSet) {
 }
 
 func setClusterOpLock(statefulSet *appsv1.StatefulSet, op SolrClusterOp) error {
+	op.LastStartTime = metav1.Now()
 	bytes, err := json.Marshal(op)
 	if err != nil {
 		return err
@@ -124,7 +125,6 @@ func retryNextQueuedClusterOp(statefulSet *appsv1.StatefulSet) (hasOp bool, err
 	hasOp = len(clusterOpRetryQueue) > 0
 	if len(clusterOpRetryQueue) > 0 {
 		nextOp := clusterOpRetryQueue[0]
-		nextOp.LastStartTime = metav1.Now()
 		err = setClusterOpLock(statefulSet, nextOp)
 		if err != nil {
 			return hasOp, err
@@ -141,7 +141,6 @@ func retryNextQueuedClusterOpWithQueue(statefulSet *appsv1.StatefulSet, clusterO
 	hasOp = len(clusterOpQueue) > 0
 	if len(clusterOpQueue) > 0 {
 		nextOp := clusterOpQueue[0]
-		nextOp.LastStartTime = metav1.Now()
 		err = setClusterOpLock(statefulSet, nextOp)
 		if err != nil {
 			return hasOp, err

controllers/solrcloud_controller.go

@@ -574,7 +574,6 @@ func (r *SolrCloudReconciler) Reconcile(ctx context.Context, req ctrl.Request) (
 			if clusterOp != nil {
 				// Starting a locked cluster operation!
 				originalStatefulSet := statefulSet.DeepCopy()
-				clusterOp.LastStartTime = metav1.Now()
 				err = setClusterOpLock(statefulSet, *clusterOp)
 				if err == nil {
 					err = r.Patch(ctx, statefulSet, client.StrategicMergeFrom(originalStatefulSet))

tests/e2e/solrcloud_rolling_upgrade_test.go

@@ -163,7 +163,7 @@ var _ = FDescribe("E2E - SolrCloud - Rolling Upgrades", func() {
 			}
 
 			By("waiting for the balanceReplicas to finish")
-			expectStatefulSetWithChecksAndTimeout(ctx, solrCloud, solrCloud.StatefulSetName(), time.Second*30, time.Second, func(g Gomega, found *appsv1.StatefulSet) {
+			expectStatefulSetWithChecksAndTimeout(ctx, solrCloud, solrCloud.StatefulSetName(), time.Second*70, time.Second, func(g Gomega, found *appsv1.StatefulSet) {
 				clusterOp, err := controllers.GetCurrentClusterOp(found)
 				g.Expect(err).ToNot(HaveOccurred(), "Error occurred while finding clusterLock for SolrCloud")
 				g.Expect(clusterOp).To(BeNil(), "StatefulSet should not have a balanceReplicas lock after balancing is complete.")

tests/e2e/suite_test.go

@@ -285,8 +285,23 @@ func writeAllSolrInfoToFiles(ctx context.Context, directory string, namespace st
 	req, err := labels.NewRequirement("technology", selection.In, []string{solrv1beta1.SolrTechnologyLabel, solrv1beta1.SolrPrometheusExporterTechnologyLabel})
 	Expect(err).ToNot(HaveOccurred())
 
-	labelSelector := labels.Everything().Add(*req)
 	listOps := &client.ListOptions{
+		Namespace: namespace,
+	}
+
+	foundSolrs := &solrv1beta1.SolrCloudList{}
+	Expect(k8sClient.List(ctx, foundSolrs, listOps)).To(Succeed(), "Could not fetch SolrClouds")
+	Expect(foundSolrs).ToNot(BeNil(), "No SolrClouds could be found")
+	for _, solrCloud := range foundSolrs.Items {
+		writeSolrClusterStatusInfoToFile(
+			ctx,
+			directory+solrCloud.Name,
+			&solrCloud,
+		)
+	}
+
+	labelSelector := labels.Everything().Add(*req)
+	listOps = &client.ListOptions{
 		Namespace:     namespace,
 		LabelSelector: labelSelector,
 	}
@@ -333,6 +348,20 @@ func writeAllSolrInfoToFiles(ctx context.Context, directory string, namespace st
 	}
 }
 
+// writeSolrClusterStatusInfoToFile writes the following each to a separate file with the given base name & directory.
+//   - SolrCloud's Cluster Status from the Collections API
+func writeSolrClusterStatusInfoToFile(ctx context.Context, baseFilename string, solrCloud *solrv1beta1.SolrCloud) {
+	clusterStatus := fetchClusterStatusWithErrorHandling(ctx, solrCloud, false)
+	if clusterStatus != "" {
+		// Write cluster status to a file
+		statusFile, err := os.Create(baseFilename + ".cluster-state.json")
+		defer statusFile.Close()
+		Expect(err).ToNot(HaveOccurred(), "Could not open file to save cluster status: %s", baseFilename+".cluster-state.json")
+		_, writeErr := statusFile.Write([]byte(clusterStatus))
+		Expect(writeErr).ToNot(HaveOccurred(), "Could not write cluster status json to file")
+	}
+}
+
 // writeAllStatefulSetInfoToFiles writes the following each to a separate file with the given base name & directory.
 //   - StatefulSet Spec/Status
 //   - StatefulSet Events

tests/e2e/test_utils_test.go

@@ -313,6 +313,10 @@ func queryCollectionWithGomega(ctx context.Context, solrCloud *solrv1beta1.SolrC
 }
 
 func fetchClusterStatus(ctx context.Context, solrCloud *solrv1beta1.SolrCloud) string {
+	return fetchClusterStatusWithErrorHandling(ctx, solrCloud, true)
+}
+
+func fetchClusterStatusWithErrorHandling(ctx context.Context, solrCloud *solrv1beta1.SolrCloud, expectNoError bool) string {
 	response, err := callSolrApiInPod(
 		ctx,
 		solrCloud,
@@ -323,7 +327,9 @@ func fetchClusterStatus(ctx context.Context, solrCloud *solrv1beta1.SolrCloud) s
 			"wt":     "json",
 		},
 	)
-	Expect(err).ToNot(HaveOccurred(), "Could not fetch clusterStatus for cloud")
+	if expectNoError {
+		Expect(err).ToNot(HaveOccurred(), "Could not fetch clusterStatus for cloud")
+	}
 
 	return response
 }