[1.13.0] - 2020-09-10
This is a large release containing fixes and features from a multitude of contributors. Some notable changes:
- Dropped support for Node 8. Given that Node 8 was EOL in the beginning of 2020, for the overwhelming majority of people this should not be an issue.
- For Typescript users this release may require you to make minor changes, due to #764, which changes the type of
KafkaMessage.valuefromBuffertoBuffer | null. This was in fact always the behavior at runtime, so this change is just fixing a bug in the types - but it will require you to handle the value beingnullfor your project to build. - Thanks to @ankon, we now support all the necessary functionality to enable Rack-Aware Follower Fetching to cut data transfer costs and reduce latency.
- Several changes to improve performance! Some larger initiatives, such as eagerly processing fetched batches in parallel, and some more contained optimizations such as reducing the amount of buffer copying during socket data buffering. We hope that these changes will have a noticeable effect on your applications, and look forward to hearing your feedback.
Added
- Add listGroup method to admin interface #645
- Add describeCluster method to admin client #648
- Add createPartitions method to admin client #661
- Add deleteGroups method to admin client #646
- Add listTopics method to admin client #718
- Add describeGroups method to admin client #742
- Allow to handle consumer retry failure at the user level #643
- Support Fetch v8 protocol (including client-side throttling) #776
- Support Fetch v9 protocol #778
- Support Fetch v10 protocol #792
- Support Fetch v11 protocol #810
- Support JoinGroup v3 and v4 protocol #801
- Oauthbearer support #680
- Add new protocol errors #824
- Add versioning to docs #835
- Add fetch topic offsets by timestamp #604
- Support LOG_APPEND_TIME record timestamps #838
- Suppress JoinGroup V4+ response error log when memberId is empty #860
Changed
- Replace fetch promise all with async generator #570
- Improve balance in the RoundRobinAssigner #635
- Add single requestTimeout runner instead of setTimeout per request #650
- Provide the subscribed topics to the protocol() function #545
- Only wait for the lock when there are enqueued batches #670
- Update consumer default retries to 5 #720 (related to #719)
- Simplify and speed up SeekOffsets #668
- Remove maxInFlight option from default retry and moved into Producer ad Consumer #754
- Use addMultipleTargetTopics instead of looping over multiple calls to addTargetTopic #748
- Only disconnect the consumer and producers if they were created #784
- Resolve socket requests without response immediately when they have been queued #785
- Move default request timeout from connection to cluster #739
- Simplify the BufferedAsyncIterator #671
- Ensure fair fetch response allocation across topic-partitions #859
Fixed
- Type improvements and fixes #636 #664 #675 #722 #729 #758 #799 #813 #757 #749 #764 #828 #843 #839
- Remove invalid topics from targetTopics on INVALID_TOPIC_EXCEPTION #666
- Fix network buffering performance problems #669
- Delete the entry for the waiter when the timeout is reached #694
- Fix encoder instanceof issue with Encoder #685
- Fix default retry for consumer #719
- Runner#waitForConsumer uses consuming stop event instead of timers #724
- Fix unhandled rejections #714 #797
- Make Array shuffle test pass in Node >= 11.0.0 #740
- Use setImmediate when scheduling calls of scheduleFetch #752
- Improve offset commit handling #775
- Use the length from the message to pre-allocate the result array #771
- Fix application lock in case of errors before the connection is established #780
- Add isNaN check for concurrency limit #787
- Fix admin client createTopics timeout #800
- Avoid repeated costly copies of buffers when working with encoders #811
- Fix fall-back retry config for producer #851