Stabilize and parallelize unit-test CI

ambry-clustermap/src/main/java/com/github/ambry/clustermap/HelixClusterManager.java

@@ -1974,9 +1974,31 @@ private void addOrUpdateInstanceInfos(Iterable<DataNodeConfig> dataNodeConfigs,
       List<ReplicaId> totalAddedReplicas = new ArrayList<>();
       List<ReplicaId> totalRemovedReplicas = new ArrayList<>();
       for (DataNodeConfig dataNodeConfig : dataNodeConfigs) {
+        String instanceName = dataNodeConfig.getInstanceName();
         Pair<List<ReplicaId>, List<ReplicaId>> addedAndRemovedReplicas;
-        if (instanceNameToAmbryDataNode.containsKey(dataNodeConfig.getInstanceName())) {
-          addedAndRemovedReplicas = updateInstanceInfo(dataNodeConfig, dcName);
+        if (instanceNameToAmbryDataNode.containsKey(instanceName)) {
+          // Update path. Mirrors createNewInstance's skip-foreign / fail-self policy: if validation
+          // throws (e.g. duplicate partition or inconsistent capacity arrived via an update to an
+          // already-known node), drop the bad node from the cluster map instead of leaving stale
+          // state behind. createNewInstance has the same wrapper inline; we keep both surfaces in
+          // sync so the skip path covers both branches uniformly.
+          try {
+            addedAndRemovedReplicas = updateInstanceInfo(dataNodeConfig, dcName);
+          } catch (Exception e) {
+            if (instanceName.equals(selfInstanceName)) {
+              logger.error(
+                  "Failed to update existing node {} (self) in datacenter {}. Failing initialization "
+                      + "since the server cannot operate with a broken local config.",
+                  instanceName, dcName, e);
+              throw e;
+            }
+            logger.error(
+                "Failed to update existing node {} in datacenter {}, removing this node from the cluster map.",
+                instanceName, dcName, e);
+            handleDataNodeDelete(instanceName);
+            dataNodeInitializationFailureCount.incrementAndGet();
+            continue;
+          }
         } else {
           addedAndRemovedReplicas = new Pair<>(createNewInstance(dataNodeConfig, dcName), new ArrayList<>());
         }

ambry-clustermap/src/test/java/com/github/ambry/clustermap/HelixClusterManagerTest.java

@@ -525,34 +525,46 @@ public void duplicatePartitionOnSameNodeSkipsNodeTest() throws Exception {
         new MockHelixCluster("AmbryTest-", testHardwareLayoutPath, testPartitionLayoutPath, testZkLayoutPath, localDc,
             useAggregatedView, 100, fullAutoCompatible ? 10000 : -1);
 
-    // Pick a node in the local DC and inject a duplicate partition across two disks in its InstanceConfig
+    // Pick a node in the local DC that has at least two disks, at least one replica to duplicate,
+    // AND is not the current server (selfInstanceName). With 3 replicas spread across N nodes some
+    // nodes have no replicas, and picking instanceConfigs.get(0) was flaky two ways: it could land
+    // on an empty node, or on the self-instance — either of which flips the test off the
+    // foreign-node skip path it intends to exercise.
     MockHelixAdmin localAdmin = testCluster.getHelixAdminFromDc(localDc);
     List<InstanceConfig> instanceConfigs = localAdmin.getInstanceConfigs("AmbryTest-" + staticClusterName);
-    InstanceConfig targetConfig = instanceConfigs.get(0);
-    String targetInstanceName = targetConfig.getInstanceName();
-
-    // Find two disk mount paths on this node and a partition on the first disk
-    Map<String, Map<String, String>> mapFields = targetConfig.getRecord().getMapFields();
+    InstanceConfig targetConfig = null;
     List<String> diskMountPaths = new ArrayList<>();
     String duplicatePartitionEntry = null;
-    for (Map.Entry<String, Map<String, String>> entry : mapFields.entrySet()) {
-      if (entry.getValue().containsKey(DISK_STATE)) {
-        diskMountPaths.add(entry.getKey());
-        if (duplicatePartitionEntry == null) {
-          String replicasStr = entry.getValue().get(REPLICAS_STR);
-          if (replicasStr != null && !replicasStr.isEmpty()) {
-            // Take the first replica entry (e.g., "0:1073741824:defaultPartitionClass,")
-            duplicatePartitionEntry = replicasStr.split(REPLICAS_DELIM_STR)[0];
+    for (InstanceConfig candidate : instanceConfigs) {
+      if (candidate.getInstanceName().equals(selfInstanceName)) {
+        continue;
+      }
+      List<String> candidateDiskPaths = new ArrayList<>();
+      String candidateReplica = null;
+      for (Map.Entry<String, Map<String, String>> entry : candidate.getRecord().getMapFields().entrySet()) {
+        if (entry.getValue().containsKey(DISK_STATE)) {
+          candidateDiskPaths.add(entry.getKey());
+          if (candidateReplica == null) {
+            String replicasStr = entry.getValue().get(REPLICAS_STR);
+            if (replicasStr != null && !replicasStr.isEmpty()) {
+              candidateReplica = replicasStr.split(REPLICAS_DELIM_STR)[0];
+            }
           }
         }
       }
+      if (candidateDiskPaths.size() >= 2 && candidateReplica != null) {
+        targetConfig = candidate;
+        diskMountPaths = candidateDiskPaths;
+        duplicatePartitionEntry = candidateReplica;
+        break;
+      }
     }
-    assertTrue("Node should have at least 2 disks", diskMountPaths.size() >= 2);
-    assertNotNull("Should find a replica to duplicate", duplicatePartitionEntry);
+    assertNotNull("No non-self instance with >=2 disks and a replica found in localDc", targetConfig);
+    String targetInstanceName = targetConfig.getInstanceName();
 
     // Add the duplicate partition to the second disk
     String secondDisk = diskMountPaths.get(1);
-    Map<String, String> secondDiskProps = mapFields.get(secondDisk);
+    Map<String, String> secondDiskProps = targetConfig.getRecord().getMapFields().get(secondDisk);
     String existingReplicas = secondDiskProps.get(REPLICAS_STR);
     secondDiskProps.put(REPLICAS_STR, existingReplicas + duplicatePartitionEntry + REPLICAS_DELIM_STR);
     localAdmin.setInstanceConfig("AmbryTest-" + staticClusterName, targetInstanceName, targetConfig);