[GitHub] [kafka] hachikuji commented on a change in pull request #9441: KAFKA-10614: Ensure group state (un)load is executed in the submitted order

[GitBox]

hachikuji commented on a change in pull request #9441:
URL: https://github.com/apache/kafka/pull/9441#discussion_r618021571



##########
File path: core/src/main/scala/kafka/coordinator/group/GroupCoordinator.scala
##########
@@ -905,19 +907,33 @@ class GroupCoordinator(val brokerId: Int,
    *
    * @param offsetTopicPartitionId The partition we are now leading
    */
-  def onElection(offsetTopicPartitionId: Int): Unit = {
-    info(s"Elected as the group coordinator for partition 
$offsetTopicPartitionId")
-    groupManager.scheduleLoadGroupAndOffsets(offsetTopicPartitionId, 
onGroupLoaded)
+  def onElection(offsetTopicPartitionId: Int, coordinatorEpoch: Int): Unit = {
+    val currentEpoch = epochForPartitionId.get(offsetTopicPartitionId)
+    if (currentEpoch.forall(currentEpoch => coordinatorEpoch > currentEpoch)) {
+      info(s"Elected as the group coordinator for partition 
$offsetTopicPartitionId in epoch $coordinatorEpoch")
+      groupManager.scheduleLoadGroupAndOffsets(offsetTopicPartitionId, 
onGroupLoaded)
+      epochForPartitionId.put(offsetTopicPartitionId, coordinatorEpoch)
+    } else {
+      warn(s"Ignored election as group coordinator for partition 
$offsetTopicPartitionId " +
+        s"in epoch $coordinatorEpoch since current epoch is $currentEpoch")
+    }
   }
 
   /**
    * Unload cached state for the given partition and stop handling requests 
for groups which map to it.
    *
    * @param offsetTopicPartitionId The partition we are no longer leading
    */
-  def onResignation(offsetTopicPartitionId: Int): Unit = {
-    info(s"Resigned as the group coordinator for partition 
$offsetTopicPartitionId")
-    groupManager.removeGroupsForPartition(offsetTopicPartitionId, 
onGroupUnloaded)
+  def onResignation(offsetTopicPartitionId: Int, coordinatorEpoch: 
Option[Int]): Unit = {
+    val currentEpoch = epochForPartitionId.get(offsetTopicPartitionId)
+    if (currentEpoch.forall(currentEpoch => currentEpoch <= 
coordinatorEpoch.getOrElse(Int.MaxValue))) {
+      info(s"Resigned as the group coordinator for partition 
$offsetTopicPartitionId in epoch $coordinatorEpoch")
+      groupManager.removeGroupsForPartition(offsetTopicPartitionId, 
onGroupUnloaded)
+      epochForPartitionId.remove(offsetTopicPartitionId)

Review comment:
       The `onResignation` hook just means we lost leadership. By keeping track 
of the epoch, we are protected from all potential reorderings.




-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org