Skip to content
This repository has been archived by the owner on Nov 20, 2020. It is now read-only.

Latest commit

 

History

History
385 lines (257 loc) · 13.5 KB

RELEASE_NOTES.md

File metadata and controls

385 lines (257 loc) · 13.5 KB

0.1.16 - 27.3.2018

  • Tested 0.1.16-alpha01

0.1.16-alpha01 - 27.3.2018

  • BUG: memory leak in 0.1.15.

0.1.15 - 19.3.2018

  • BREAKING: FetchResponse item now explicit structure rather than tuple. Breaking only if using low-level Fetch API.

  • BUG: consumer doesn't recover from leaderless_partition_detected #204

  • BUG: consumer would enter an infinite loop between offsets_out_of_range and resuming_fetch_from_reset_offsets

  • BUG: stalled consumers after TCP connection timeout

  • BUG: offsets committed for partitions which are no longer assigned after a rebalance

  • FEATURE: ConsumerConfig.fetchMaxBytesOverride allows increasing fetchMaxBytes to a ceiling on message too big exceptions.

  • FEATURE: ProducerQueueType to allow configuration of producer message queue networks.

  • IMPROVE: support for v0.11 of broker protocol, including v5 of Fetch and v3 of Produce APIs

  • IMPROVE: logging where received_empty_buffer logged as INFO message as it doesn't indicate an error

  • IMPROVE: connection managemeng to ensure only a single TCP connection per broker per Kafka connection

  • IMPROVE: Reduce recovery time from broker restarts and rebalancing

  • IMPROVE: improved TCP connection logging and log levels to reduce "false positives"

  • IMPROVE: .NET Standard support

0.1.15-alpha05 - 19.3.2018

  • BUG: ensure CRC check runs with respect to appropriate message version (magic byte)

0.1.15-alpha04 - 19.3.2018

  • BUG: compression works when using v0.11 (auto) protocol

0.1.15-alpha02 - 14.3.2018

  • IMPROVE: support for v0.11 of broker protocol, including v5 of Fetch and v3 of Produce APIs
  • IMPROVE: improved TCP connection logging and log levels to reduce "false positives"
  • BREAKING: FetchResponse item now explicit structure rather than tuple. Breaking only if using low-level Fetch API.
  • FEATURE: ProducerQueueType to allow configuration of producer message queue networks.
  • BUG: consumer doesn't recover from leaderless_partition_detected #204

0.1.15-alpha01 - 5.3.2018

  • BUG: consumer would enter an infinite loop between offsets_out_of_range and resuming_fetch_from_reset_offsets
  • BUG: stalled consumers after TCP connection timeout
  • BUG: offsets committed for partitions which are no longer assigned after a rebalance
  • FEATURE: ConsumerConfig.fetchMaxBytesOverride allows increasing fetchMaxBytes to a ceiling on message too big exceptions.
  • IMPROVE: logging where received_empty_buffer logged as INFO message as it doesn't indicate an error
  • IMPROVE: connection managemeng to ensure only a single TCP connection per broker per Kafka connection
  • IMPROVE: Reduce recovery time from broker restarts and rebalancing

0.1.14-alpha02 - 28.2.2018

  • Adjusting FSharp.Core reference

0.1.14-alpha01 - 27.12.2017

  • .NET Standard + .NET 45 targets

0.1.13 - 20.12.2017

  • Roll-up of 0.1.13-alpha builds after testing

0.1.13-alpha02 - 19.12.2017

  • WARN: Default to use v0.10.1 of broker protocol.
  • Print client and protocol version.

0.1.13-alpha01 - 18.12.2017

  • Consumers will continue to run if not assigned partitions.
  • Fixed 'Consumer.fetchOffsets' returning no offsets when Array.empty is passed in as partitions per topic
  • Fixed 'Consumer.stream' function that sometimes causes starvation when AsyncSeq.mergeAll would prioritize first few tasks rather than later
  • Added auto versioning support for Kafka APIs for Produce, Fetch, Offset, Metadata, OffsetCommit, OffsetFetch, GroupCoordinator, JoinGroup, Heartbeat, LeaveGroup, SyncGroup, DescribeGroups, ListGroups, ApiVersions.
  • Added an INFO message printed by connection to show protocol and client versions.
  • Adjusted ConsumerConfig.[fetchMaxWaitMs, fetchMinBytes] to match Kafka native client defaults.
  • BufferingProducer.produceBatched

0.1.12 - 13.12.2017

  • Buffering producer experimental module.

0.1.11 - 12.12.2017

  • Added support for all versions of the OffsetFetch API in the Kafka protocol.
  • BREAKING: Offset commit queue is moved to module PeriodicCommitQueue. See: https://jet.github.io/kafunk/consuming.html#Periodic-Offset-Commit.
  • Producer performance improvements.
  • Fixed bug with producing using version >= 0.10.0 wherein empty key/value would be written

0.1.10 - 15.11.2017

  • Fix another scenario referenced in 0.1.9 producer batching fix.

0.1.9 - 13.11.2017

  • Fix Producer batching wherein if Producer.produceBatched was called with a batch size larger than the max, it would still send it.

0.1.8 - 30.10.2017

  • Fix producer recovery wherein a produce batch timeout may overlap with broker connection failures, resulting in retries without rediscovery.

0.1.7 - 24.10.2017

  • Fix Join/Sync group response for v10 and greater

0.1.6 - 23.10.2017

  • Cumulitive release.

0.1.6-rc3 - 18.10.2017

  • NEW: ConsumerInfo.consumerGroups returns information about all consumer groups in the cluster, including group members and the topics-partitions they are assgined.

0.1.6-rc2 - 17.10.2017

  • Producer retry without recovery when recovery not needed.
  • Ensure that Resource isn't corrupted after a failed resource recovery attempt and can be retried again.
  • Resource performance improvement.

0.1.6-rc1 - 13.10.2017

  • Fixed 'ConsumerProgress.progress' returning -1 for offsets when Array.empty is passed in

0.1.5 - 11.10.2017

  • Cumulitive release with fixes since 0.1.4

0.1.5-rc6 - 10.10.2017

  • Producer performance improvement

0.1.5-rc5 - 07.10.2017

  • BUG: Producer.produceBatched doesn't recover after transient failures

0.1.5-rc4 - 06.10.2017

  • BUG: race condition in consumer group join wherein multiple groups created on the same connection might prevent a group coordinator from being retrieved

0.1.5-rc3 - 02.10.2017

  • Adjust retry policy defaults: reduce retries to fail fast
  • Exposed internal ConsumerGroup.decodeMemberAssignment
  • Improve consumer group stability, ensuring escalation in all consumer failure modes

0.1.5-rc2 - 27.09.2017

  • AsyncSeq v2.0.14
  • Change offset range argument type Consumer.streamRange

0.1.5-rc1 - 26.09.2017

  • Added Consumer.streamRange to read a specified range of offsets.
  • Added Consumer.periodicOffsetCommitter which can help ensuring periodic offset commits when using Consumer.stream.
  • Logging improvements: changed some ERROR messages to WARN, added group coordinator to group related logs, caught missing partition info error
  • AsyncSeq v2.0.13

0.1.4 - 03.08.2017

  • Fixed FetchResponse MessageTooBigException when a message set has been compressed. (#160)

0.1.4-beta - 25.07.2017

  • Fixed v0.10.1 protocol bug for Offset API.
  • Fixed v0.10.1 protocol bug for JoinGroup API.
  • Added ApiVersionsRequest.
  • Added support for automatically detecting API versions via ApiVersionRequest using KafkaConfig.autoApiVersions.

0.1.3-alpha - 14.07.2017

  • Added ProducerResult.count indicating the number of messages produced in a batch for a partition.

0.1.2-alpha - 03.07.2017

  • Fixed consumer offset commit bug wherein after a rebalance a consumer gets assigned a new partition which doesn't receive any messages for longer than the offset retention period, the offsets would be lost. This would only happen after a rebalance not after initial join.

0.1.1-alpha - 25.05.2017

  • Snappy compression.
  • Fixed lag calculation bug in ConsumerInfo module.

0.1.0 - 01.05.2017

  • Ensure recovery during broker restart (virtual broker routing).
  • Add Producer.produceBatched with improved API and performance:
    • Takes a batch of messages rather than a function returning a batch.
    • Parallelizes requests across brokers.

0.0.43-alpha001 - 24.04.2017

  • Fix message set decompression (take 3)

0.0.41-alpha001 - 24.04.2017

  • Fix message set decompression (take 2)

0.0.40-alpha001 - 21.04.2017

  • Fix message set decompression

0.0.39-alpha001 - 11.04.2017

  • Unroll loop in producer to make more efficient
  • Request buffer pool
  • Added ZK offset migration script in tests project
  • Support v10.0.1 for Offset API
  • BREAKING:
    • MessageSet tuples replaced with structs
    • ProduceRequest tuples replaced with structs
    • ProduceResponse tuples replaced with structs
    • OffsetCommitRequest added support for v1 (in addition to v0 and v2)

0.0.38-alpha001 - 29.03.2017

  • Fix use of IVar where concurrent puts are possible
  • Fix #124
  • Fix #126

0.0.37-alpha001 - 27.03.2017

  • Fix producer bug where during recovery, messages in flight would be lost and never timeout.

0.0.36-alpha001 - 23.03.2017

  • Improve produce and fetch codec performance.
  • Fix possible deadlock with Async.AwaitTask usage.

0.0.35-alpha001 - 08.04.2017

  • Fix CRC32 check.

0.0.34-alpha001 - 08.03.2017

  • Make fetch response decoding more efficient (eliminate intermediate tuple allocations).
  • Make CRC check on fetch response configurable (defaults to true).

0.0.33-alpha001 - 01.03.2017

  • Ensure fetch errors are escalated to the top-level consumer.

0.0.32-alpha001 - 28.02.2017

  • Fix bug in ConsumerInfo module where offset information for all partitions wouldn't be retrieved and would throw during the merge.
  • Fix Async.parallel used internall for testing.

0.0.31-alpha001 - 24.02.2017

  • Allow disabling of Console logger
  • Ensure periodic offset committer commits offsets on start

0.0.30-alpha001 - 24.02.2017

  • Improve producer batching performance
  • Fix all offset commit

0.0.29-alpha001 - 23.02.2017

  • Improve producer performance
  • Improve producer error messaging

0.0.28-alpha001 - 22.02.2017

  • Commit offsets even when unchanged to prevent loss due to retention.
  • BREAKING:
    • Removed ConsumerConfig.initialFetchTime, consolidated into ConsumerConfig.autoOffsetReset
    • Replaced offsetOutOfRangeAction with autoOffsetReset and new union type AutoOffsetReset

0.0.27-alpha001 - 16.02.2017

  • Hide internal members, including Async
  • Refine fault tolerance defaults.

0.0.26-alpha001 - 14.02.2017

  • Special Valentine's day edition.
  • Refined flow for handling escalated failures.
  • Fix error code MessageSizeTooLarge.
  • Default client.id = "" rather than Guid; new Guid for connection id.
  • Expose connection, producer, consumer config.

0.0.25-alpha001 - 10.02.2017

  • Default producer to RequiredAcks.AllInSync
  • Fix range consumer partition assignment strategy

0.0.24-alpha001 - 08.02.2017

  • Adjust consumer group heartbeat defaults
  • Added AsyncSeq tests

0.0.23-alpha001 - 08.02.2017

  • Improve producer performance (batch size measurement)

0.0.22-alpha001 - 08.02.2017

  • Producer in-memory buffer in bytes
  • ConsumerInfo module for consumer progress tracking

0.0.21-alpha001 - 07.02.2017

  • Updated exampled/readme
  • Breaking changes:
    • Consumer.stream drops bufferSize parameter
    • ProducerResult contains a single partition-offset pair rather than array

0.0.20-alpha001 - 06.02.2017

  • Producer buffering by size in bytes
  • Fixex bug where producer errors weren't being surfaced
  • Breaking changes:
    • ProducerConfig buffer size settings are new
    • Producer.produce takes a single message rather than array
    • Consumer.consume arguments reordered
  • Logging improvements (more compact, more info)
  • Improved producer-consumer test

0.0.19-alpha001 - 25.01.2017

  • Producer buffering
  • Ensure cancellation propagated to consumer on group close
  • Logging improvements

0.0.18-alpha001 - 20.01.2017

  • Fix ConsumerMessageSet.lag measure
  • Propagate cancellation to consumer fetch process

0.0.17-alpha001 - 17.01.2017

  • Discovery brokers by DNS when appropriate
  • Explicit channel failure and recovery contract

0.0.16-alpha001 - 09.01.2017

  • Fix bug in assignment strategy where all available metadate were used rather than that of subscribed topic

0.0.15-alpha001 - 09.01.2017

  • Pass consumer state to consume callback (BREAKING)

0.0.14-alpha001 - 09.01.2017

  • Consumer group assignment strategies configurable

0.0.13-alpha001 - 06.01.2017

  • Refine Fetch consumer offsets API

0.0.12-alpha001 - 06.01.2017

  • Current consumer state API
  • Fetch consumer offsets API

0.0.11-alpha001 - 05.01.2017

  • Hide internal members
  • Documentation

0.0.10-alpha001 - 04.01.2017

  • Log leader-size member assignments
  • Fix consumer NotCoordinatorForGroup error recovery

0.0.9-alpha001 - 03.01.2017

  • Refine TCP error recovery

0.0.8-alpha001 - 03.01.2017

  • v0.10.1 support

0.0.7-alpha001 - 02.01.2017

  • Fix overflow bug in Partitioner.roundRobin
  • Error flow monitoring

0.0.6 - 30.12.2016

  • Fix bug in Async.choose
  • Consumer.commitOffsetsToTime
  • Consumer doesn't fetch offsets until its starts consuming

0.0.5 - 30.12.2016

  • Auto recover producer from error code 2

0.0.4 - 29.12.2016

  • Consumer fetch request batching by broker
  • Fix protocol codec issue for batched fetch responses
  • Periodic offset commit

0.0.3 - 28.12.2016

  • Fix a few reconnection/fault tolerance bugs
  • Adjust consumer api to allow offset strategies to be parameterized

0.0.2 - 29.11.2016

  • A more complete consumer API is now available
  • We now use FSharp.Control.AsyncSeq instead of an internal version
  • Adjustments have been made to connection recovery and general fault tolerance (still early)
  • Improvements to message grouping should improve producer throughput

0.0.1 - 20.07.2016

  • initial