jsancio commented on a change in pull request #11733:
URL: https://github.com/apache/kafka/pull/11733#discussion_r813995003
##########
File path:
metadata/src/main/java/org/apache/kafka/controller/ReplicationControlManager.java
##########
@@ -760,44 +719,132 @@ BrokersToIsrs brokersToIsrs() {
}
if (change.leader() != request.brokerId() &&
change.leader() != NO_LEADER_CHANGE) {
- // Normally, an alterIsr request, which is made by the
partition
+ // Normally, an alterPartition request, which is made
by the partition
// leader itself, is not allowed to modify the
partition leader.
// However, if there is an ongoing partition
reassignment and the
// ISR change completes it, then the leader may change
as part of
// the changes made during reassignment cleanup.
//
// In this case, we report back FENCED_LEADER_EPOCH to
the leader
- // which made the alterIsr request. This lets it know
that it must
+ // which made the alterPartition request. This lets it
know that it must
// fetch new metadata before trying again. This return
code is
// unusual because we both return an error and
generate a new
// metadata record. We usually only do one or the
other.
- log.info("AlterIsr request from node {} for {}-{}
completed " +
+ log.info("AlterPartition request from node {} for
{}-{} completed " +
"the ongoing partition reassignment and triggered
a " +
- "leadership change. Reutrning
FENCED_LEADER_EPOCH.",
+ "leadership change. Returning
FENCED_LEADER_EPOCH.",
request.brokerId(), topic.name, partitionId);
- responseTopicData.partitions().add(new
AlterIsrResponseData.PartitionData().
+ responseTopicData.partitions().add(new
AlterPartitionResponseData.PartitionData().
setPartitionIndex(partitionId).
setErrorCode(FENCED_LEADER_EPOCH.code()));
continue;
} else if (change.removingReplicas() != null ||
change.addingReplicas() != null) {
- log.info("AlterIsr request from node {} for {}-{}
completed " +
+ log.info("AlterPartition request from node {} for
{}-{} completed " +
"the ongoing partition reassignment.",
request.brokerId(),
topic.name, partitionId);
}
}
- responseTopicData.partitions().add(new
AlterIsrResponseData.PartitionData().
+
+ responseTopicData.partitions().add(new
AlterPartitionResponseData.PartitionData().
setPartitionIndex(partitionId).
setErrorCode(result.code()).
setLeaderId(partition.leader).
+ setIsr(Replicas.toList(partition.isr)).
+
setLeaderRecoveryState(partition.leaderRecoveryState.value()).
Review comment:
It is a tagged field so we can't. I think this is okay for the following
reason:
1. If the leader doesn't support this feature it will send version 0 of the
AlterPartition request which will set the leader recovery state to RECOVERED
2. The leader only change value if:
a. a AlterPartition request is sent that increases the ISR
b. another unclean leader election is performed which sets (keeps) the
leader recovery state as RECOVERING
Note that the broker to controller channel uses the ApiVersions response to
determine which version of AlterPartition to send.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
