[ https://issues.apache.org/jira/browse/KAFKA-9837?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17487938#comment-17487938 ]
dengziming commented on KAFKA-9837: ----------------------------------- Thank you [~cmccabe] , your concern make sense to me. it seems o(num_partitions_in_dir) RPCs is inevitable here since the controller will write o(n) `PartitionChangeRecord` in Kraft metadata topic and send to raft observers(in zk mode, controller will send o(n) `UpdateMetadataRequest` and `StopReplicaRequest`). > New RPC for notifying controller of failed replica > -------------------------------------------------- > > Key: KAFKA-9837 > URL: https://issues.apache.org/jira/browse/KAFKA-9837 > Project: Kafka > Issue Type: New Feature > Components: controller, core > Reporter: David Arthur > Assignee: dengziming > Priority: Major > Labels: kip-500 > Fix For: 3.2.0 > > > This is the tracking ticket for > [KIP-589|https://cwiki.apache.org/confluence/display/KAFKA/KIP-589+Add+API+to+update+Replica+state+in+Controller]. > For the bridge release, brokers should no longer use ZooKeeper to notify the > controller that a log dir has failed. It should instead use an RPC mechanism. -- This message was sent by Atlassian Jira (v8.20.1#820001)