KAFKA-15364: Replay BrokerRegistrationChangeRecord.logDirs (#14998)

Any directory changes must be considered when replaying
BrokerRegistrationChangeRecord. This is necessary
to persist directory failures in the cluster metadata,
which #14902 missed.

Reviewers: Omnia G.H Ibrahim <o.g.h.ibrahim@gmail.com>, Viktor Somogyi-Vass <viktorsomogyi@gmail.com>

metadata/src/main/java/org/apache/kafka/controller/ClusterControlManager.java

@@ -534,7 +534,8 @@ public class ClusterControlManager {
             record.id(),
             record.epoch(),
             BrokerRegistrationFencingChange.FENCE.asBoolean(),
-            BrokerRegistrationInControlledShutdownChange.NONE.asBoolean()
+            BrokerRegistrationInControlledShutdownChange.NONE.asBoolean(),
+            Optional.empty()
         );
     }
 
@@ -544,7 +545,8 @@ public class ClusterControlManager {
             record.id(),
             record.epoch(),
             BrokerRegistrationFencingChange.UNFENCE.asBoolean(),
-            BrokerRegistrationInControlledShutdownChange.NONE.asBoolean()
+            BrokerRegistrationInControlledShutdownChange.NONE.asBoolean(),
+            Optional.empty()
         );
     }
 
@@ -557,12 +559,14 @@ public class ClusterControlManager {
             BrokerRegistrationInControlledShutdownChange.fromValue(record.inControlledShutdown()).orElseThrow(
                 () -> new IllegalStateException(String.format("Unable to replay %s: unknown " +
                     "value for inControlledShutdown field: %x", record, record.inControlledShutdown())));
+        Optional<List<Uuid>> directoriesChange = Optional.ofNullable(record.logDirs()).filter(list -> !list.isEmpty());
         replayRegistrationChange(
             record,
             record.brokerId(),
             record.brokerEpoch(),
             fencingChange.asBoolean(),
-            inControlledShutdownChange.asBoolean()
+            inControlledShutdownChange.asBoolean(),
+            directoriesChange
         );
     }
 
@@ -571,7 +575,8 @@ public class ClusterControlManager {
         int brokerId,
         long brokerEpoch,
         Optional<Boolean> fencingChange,
-        Optional<Boolean> inControlledShutdownChange
+        Optional<Boolean> inControlledShutdownChange,
+        Optional<List<Uuid>> directoriesChange
     ) {
         BrokerRegistration curRegistration = brokerRegistrations.get(brokerId);
         if (curRegistration == null) {
@@ -583,7 +588,8 @@ public class ClusterControlManager {
         } else {
             BrokerRegistration nextRegistration = curRegistration.cloneWith(
                 fencingChange,
-                inControlledShutdownChange
+                inControlledShutdownChange,
+                directoriesChange
             );
             if (!curRegistration.equals(nextRegistration)) {
                 log.info("Replayed {} modifying the registration for broker {}: {}",

metadata/src/main/java/org/apache/kafka/controller/ReplicationControlManager.java

@@ -1401,10 +1401,13 @@ public class ReplicationControlManager {
                         "handleDirectoriesOffline[" + brokerId + ":" + newOfflineDir + "]",
                         brokerId, NO_LEADER, records, iterator);
             }
+            List<Uuid> newOnlineDirs = registration.directoryDifference(offlineDirs);
             records.add(new ApiMessageAndVersion(new BrokerRegistrationChangeRecord().
                     setBrokerId(brokerId).setBrokerEpoch(brokerEpoch).
-                    setLogDirs(registration.directoryDifference(offlineDirs)),
+                    setLogDirs(newOnlineDirs),
                     (short) 2));
+            log.warn("Directories {} in broker {} marked offline, remaining directories: {}",
+                    newOfflineDirs, brokerId, newOnlineDirs);
         }
     }
 
@@ -2126,6 +2129,11 @@ public class ReplicationControlManager {
                             if (directoryIsOffline) {
                                 leaderAndIsrUpdates.add(new TopicIdPartition(topicId, partitionIndex));
                             }
+                            if (log.isDebugEnabled()) {
+                                log.debug("Broker {} assigned partition {}:{} to {} dir {}",
+                                    brokerId, topics.get(topicId).name(), partitionIndex,
+                                    directoryIsOffline ? "OFFLINE" : "ONLINE", dirId);
+                            }
                         }
                     }
                     resTopic.partitions().add(new AssignReplicasToDirsResponseData.PartitionData().

metadata/src/main/java/org/apache/kafka/image/ClusterDelta.java

@@ -17,6 +17,7 @@
 
 package org.apache.kafka.image;
 
+import org.apache.kafka.common.Uuid;
 import org.apache.kafka.common.metadata.BrokerRegistrationChangeRecord;
 import org.apache.kafka.common.metadata.FenceBrokerRecord;
 import org.apache.kafka.common.metadata.RegisterBrokerRecord;
@@ -30,6 +31,7 @@ import org.apache.kafka.metadata.ControllerRegistration;
 import org.apache.kafka.server.common.MetadataVersion;
 
 import java.util.HashMap;
+import java.util.List;
 import java.util.Map;
 import java.util.Map.Entry;
 import java.util.Optional;
@@ -112,6 +114,7 @@ public final class ClusterDelta {
         BrokerRegistration curRegistration = getBrokerOrThrow(record.id(), record.epoch(), "fence");
         changedBrokers.put(record.id(), Optional.of(curRegistration.cloneWith(
             BrokerRegistrationFencingChange.FENCE.asBoolean(),
+            Optional.empty(),
             Optional.empty()
         )));
     }
@@ -120,6 +123,7 @@ public final class ClusterDelta {
         BrokerRegistration curRegistration = getBrokerOrThrow(record.id(), record.epoch(), "unfence");
         changedBrokers.put(record.id(), Optional.of(curRegistration.cloneWith(
             BrokerRegistrationFencingChange.UNFENCE.asBoolean(),
+            Optional.empty(),
             Optional.empty()
         )));
     }
@@ -135,9 +139,11 @@ public final class ClusterDelta {
             BrokerRegistrationInControlledShutdownChange.fromValue(record.inControlledShutdown()).orElseThrow(
                 () -> new IllegalStateException(String.format("Unable to replay %s: unknown " +
                     "value for inControlledShutdown field: %d", record, record.inControlledShutdown())));
+        Optional<List<Uuid>> directoriesChange = Optional.ofNullable(record.logDirs()).filter(list -> !list.isEmpty());
         BrokerRegistration nextRegistration = curRegistration.cloneWith(
             fencingChange.asBoolean(),
-            inControlledShutdownChange.asBoolean()
+            inControlledShutdownChange.asBoolean(),
+            directoriesChange
         );
         if (!curRegistration.equals(nextRegistration)) {
             changedBrokers.put(record.brokerId(), Optional.of(nextRegistration));

metadata/src/main/java/org/apache/kafka/metadata/BrokerRegistration.java

@@ -390,12 +390,14 @@ public class BrokerRegistration {
 
     public BrokerRegistration cloneWith(
         Optional<Boolean> fencingChange,
-        Optional<Boolean> inControlledShutdownChange
+        Optional<Boolean> inControlledShutdownChange,
+        Optional<List<Uuid>> directoriesChange
     ) {
         boolean newFenced = fencingChange.orElse(fenced);
         boolean newInControlledShutdownChange = inControlledShutdownChange.orElse(inControlledShutdown);
+        List<Uuid> newDirectories = directoriesChange.orElse(directories);
 
-        if (newFenced == fenced && newInControlledShutdownChange == inControlledShutdown)
+        if (newFenced == fenced && newInControlledShutdownChange == inControlledShutdown && newDirectories.equals(directories))
             return this;
 
         return new BrokerRegistration(
@@ -408,7 +410,7 @@ public class BrokerRegistration {
             newFenced,
             newInControlledShutdownChange,
             isMigratingZkBroker,
-            directories
+            newDirectories
         );
     }
 }

metadata/src/test/java/org/apache/kafka/controller/ReplicationControlManagerTest.java

@@ -2979,6 +2979,8 @@ public class ReplicationControlManagerTest {
             sortPartitionChangeRecords(filter(records, PartitionChangeRecord.class))
         );
         assertEquals(3, records.size());
+        ctx.replay(records);
+        assertEquals(Collections.singletonList(dir2b1), ctx.clusterControl.registration(b1).directories());
     }
 
     /**