KAFKA-12339: Add retry to admin client's listOffsets (#10152)

chia7712 · rhauch · commit fe132ee29329 · 2021-02-19T18:05:10.000-06:00
`KafkaAdmin.listOffsets` did not handle topic-level errors, hence the UnknownTopicOrPartitionException on topic-level can obstruct a Connect worker from running when the new internal topic is NOT synced to all brokers. The method did handle partition-level retriable errors by retrying, so this changes to handle topic-level retriable errors in the same way.

This allows a Connect worker to start up and have the admin client retry when the worker is trying to read to the end of the newly-created internal topics until the internal topic metadata is synced to all brokers.

Author: Chia-Ping Tsai &lt;chia7712@gmail.com&gt;
Reviewers: Randall Hauch &lt;rhauch@gmail.com&gt;, Konstantine Karantasis &lt;konstantine@confluent.io&gt;
diff --git a/clients/src/main/java/org/apache/kafka/clients/admin/internals/MetadataOperationContext.java b/clients/src/main/java/org/apache/kafka/clients/admin/internals/MetadataOperationContext.java
@@ -82,6 +82,7 @@ public Collection<String> topics() {
 
     public static void handleMetadataErrors(MetadataResponse response) {
         for (TopicMetadata tm : response.topicMetadata()) {
+            if (shouldRefreshMetadata(tm.error())) throw tm.error().exception();
             for (PartitionMetadata pm : tm.partitionMetadata()) {
                 if (shouldRefreshMetadata(pm.error)) {
                     throw pm.error.exception();
diff --git a/clients/src/test/java/org/apache/kafka/clients/admin/KafkaAdminClientTest.java b/clients/src/test/java/org/apache/kafka/clients/admin/KafkaAdminClientTest.java
@@ -342,11 +342,15 @@ private static FindCoordinatorResponse prepareFindCoordinatorResponse(Errors err
     }
 
     private static MetadataResponse prepareMetadataResponse(Cluster cluster, Errors error) {
+        return prepareMetadataResponse(cluster, error, error);
+    }
+
+    private static MetadataResponse prepareMetadataResponse(Cluster cluster, Errors topicError, Errors partitionError) {
         List<TopicMetadata> metadata = new ArrayList<>();
         for (String topic : cluster.topics()) {
             List<PartitionMetadata> pms = new ArrayList<>();
             for (PartitionInfo pInfo : cluster.availablePartitionsForTopic(topic)) {
-                PartitionMetadata pm = new PartitionMetadata(error,
+                PartitionMetadata pm = new PartitionMetadata(partitionError,
                         new TopicPartition(topic, pInfo.partition()),
                         Optional.of(pInfo.leader().id()),
                         Optional.of(234),
@@ -355,7 +359,7 @@ private static MetadataResponse prepareMetadataResponse(Cluster cluster, Errors
                         Arrays.stream(pInfo.offlineReplicas()).map(Node::id).collect(Collectors.toList()));
                 pms.add(pm);
             }
-            TopicMetadata tm = new TopicMetadata(error, topic, false, pms);
+            TopicMetadata tm = new TopicMetadata(topicError, topic, false, pms);
             metadata.add(tm);
         }
         return MetadataResponse.prepareResponse(0,
@@ -2342,6 +2346,39 @@ public void testListOffsets() throws Exception {
         }
     }
 
+    @Test
+    public void testListOffsetsRetriableErrorOnMetadata() throws Exception {
+        Node node = new Node(0, "localhost", 8120);
+        List<Node> nodes = Collections.singletonList(node);
+        final Cluster cluster = new Cluster(
+            "mockClusterId",
+            nodes,
+            Collections.singleton(new PartitionInfo("foo", 0, node, new Node[]{node}, new Node[]{node})),
+            Collections.emptySet(),
+            Collections.emptySet(),
+            node);
+        final TopicPartition tp0 = new TopicPartition("foo", 0);
+
+        try (AdminClientUnitTestEnv env = new AdminClientUnitTestEnv(cluster)) {
+            env.kafkaClient().setNodeApiVersions(NodeApiVersions.create());
+            env.kafkaClient().prepareResponse(prepareMetadataResponse(cluster, Errors.UNKNOWN_TOPIC_OR_PARTITION, Errors.NONE));
+            // metadata refresh because of UNKNOWN_TOPIC_OR_PARTITION
+            env.kafkaClient().prepareResponse(prepareMetadataResponse(cluster, Errors.NONE));
+            // listoffsets response from broker 0
+            Map<TopicPartition, PartitionData> responseData = new HashMap<>();
+            responseData.put(tp0, new PartitionData(Errors.NONE, -1L, 123L, Optional.of(321)));
+            env.kafkaClient().prepareResponse(new ListOffsetResponse(responseData));
+
+            ListOffsetsResult result = env.adminClient().listOffsets(Collections.singletonMap(tp0, OffsetSpec.latest()));
+
+            Map<TopicPartition, ListOffsetsResultInfo> offsets = result.all().get(3, TimeUnit.SECONDS);
+            assertEquals(1, offsets.size());
+            assertEquals(123L, offsets.get(tp0).offset());
+            assertEquals(321, offsets.get(tp0).leaderEpoch().get().intValue());
+            assertEquals(-1L, offsets.get(tp0).timestamp());
+        }
+    }
+
     @Test
     public void testListOffsetsRetriableErrors() throws Exception {
 
