[KAFKA-12258] Change the BatchAccumulator to split records into batches

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.8.0
• Component/s: controller, core
• Labels: kip-500
• Assignee: Jose Armando Garcia Sancio
• Reporter: Alok Nikhil
• Created: 30/Jan/21 00:52
• Updated: 19/Feb/21 03:46
• Resolved: 19/Feb/21 03:46

Description

Modify the `BatchAccumulator.append contract` to support splitting a batch of records whose size is greater than the maximum allowed size (of 1048576 currently) into batches to avoid RaftClient failures such as

leader=0, leaderEpoch=0, partitionEpoch=0) at version 0), ApiMessageAndVersion(TopicRecord(name='topic-BEHRW9999', topicId=6cRudOGO3yqlsu48RwyPSw) at version 0), ApiMessageAndVersion(PartitionRecord(partitionId=0, topicId=6cRudOGO3yqlsu48RwyPSw, replicas=[1, 2, 0], isr=[1, 2, 0], removingReplicas=null, addingReplicas=null, leader=1, leaderEpoch=0, partitionEpoch=0) at version 0)] is 1088890, which exceeds the maximum allowed batch size of 1048576 Jan 30 00:13:40 ip-10-0-0-254 kafka-server-start.sh[633637]: at org.apache.kafka.raft.internals.BatchAccumulator.append(BatchAccumulator.java:110) Jan 30 00:13:40 ip-10-0-0-254 kafka-server-start.sh[633637]: at org.apache.kafka.raft.KafkaRaftClient.scheduleAppend(KafkaRaftClient.java:1885) Jan 30 00:13:40 ip-10-0-0-254 kafka-server-start.sh[633637]: at org.apache.kafka.raft.metadata.MetaLogRaftShim.scheduleWrite(MetaLogRaftShim.java:60) Jan 30 00:13:40 ip-10-0-0-254 kafka-server-start.sh[633637]: at org.apache.kafka.controller.QuorumController$ControllerWriteEvent.run(QuorumController.java:406) Jan 30 00:13:40 ip-10-0-0-254 kafka-server-start.sh[633637]: at org.apache.kafka.common.utils.KafkaEventQueue$EventContext.run(KafkaEventQueue.java:117) Jan 30 00:13:40 ip-10-0-0-254 kafka-server-start.sh[633637]: at org.apache.kafka.common.utils.KafkaEventQueue$EventHandler.handleEvents(KafkaEventQueue.java:192) Jan 30 00:13:40 ip-10-0-0-254 kafka-server-start.sh[633637]: at org.apache.kafka.common.utils.KafkaEventQueue$EventHandler.run(KafkaEventQueue.java:165) Jan 30 00:13:40 ip-10-0-0-254 kafka-server-start.sh[633637]: at java.base/java.lang.Thread.run(Thread.java:834) Jan 30 00:13:40 ip-10-0-0-254 kafka-server-start.sh[633637]: [2021-01-30 00:13:40,277] INFO [Controller 3000] Reverting to snapshot 2232 (org.apache.kafka.timeline.SnapshotRegistry)

Example use-case: Creating 10,000 topics in a single API call

Issue Links

• links to: GitHub Pull Request #10063