GitHub user onurkaraman opened a pull request:
https://github.com/apache/kafka/pull/3765
KAFKA-5642 [WIP]: Use async ZookeeperClient in Controller
Synchronous zookeeper writes means that we wait an entire round trip before
doing the next write. These synchronous writes are happening at a per-partition
granularity in several places, so partition-heavy clusters suffer from the
controller doing many sequential round trips to zookeeper:
- PartitionStateMachine.electLeaderForPartition updates leaderAndIsr in
zookeeper on transition to OnlinePartition. This gets triggered per-partition
sequentially with synchronous writes during controlled shutdown of the shutting
down broker's replicas for which it is the leader.
- ReplicaStateMachine updates leaderAndIsr in zookeeper on transition to
OfflineReplica when calling KafkaController.removeReplicaFromIsr. This gets
triggered per-partition sequentially with synchronous writes for failed or
controlled shutdown brokers.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/onurkaraman/kafka KAFKA-5642
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/kafka/pull/3765.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #3765
----
commit 4b3e4e8b1d403bf817aeaa9196fb69a15b1a13a0
Author: Onur Karaman <[email protected]>
Date: 2017-07-27T18:29:35Z
KAFKA-5642: Use async ZookeeperClient in Controller
Synchronous zookeeper writes means that we wait an entire round trip before
doing the next write. These synchronous writes are happening at a per-partition
granularity in several places, so partition-heavy clusters suffer from the
controller doing many sequential round trips to zookeeper:
- PartitionStateMachine.electLeaderForPartition updates leaderAndIsr in
zookeeper on transition to OnlinePartition. This gets triggered per-partition
sequentially with synchronous writes during controlled shutdown of the shutting
down broker's replicas for which it is the leader.
- ReplicaStateMachine updates leaderAndIsr in zookeeper on transition to
OfflineReplica when calling KafkaController.removeReplicaFromIsr. This gets
triggered per-partition sequentially with synchronous writes for failed or
controlled shutdown brokers.
----
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
