Apache Flink 1.13.6 发版公告

本文翻译自 Apache Flink 社区博文——《Apache Flink 1.13.6 Release Announcement》
。

Apache Flink 社区很荣幸地宣布 Flink 1.13 的另一个错误修复版本正式发布。

此版本包括 99 个错误和漏洞修复以及 Flink 1.13 的小改进，包括升级 Apache Log4j 版本至 2.17.1。您可在下面查看所有错误修复和改进的列表（不包括对构建基础架构和构建稳定性的改进）。有关所有更改的完整列表，请参阅：JIRA 列表。

我们强烈建议所有用户升级到 Flink 1.13.6 版本。

发行工件（Release Artifacts）

Maven 依赖（Maven Dependencies）

<dependency>
  <groupId>org.apache.flink</groupId>
  <artifactId>flink-java</artifactId>
  <version>1.13.6</version>
</dependency>
<dependency>
  <groupId>org.apache.flink</groupId>
  <artifactId>flink-streaming-java_2.11</artifactId>
  <version>1.13.6</version>
</dependency>
<dependency>
  <groupId>org.apache.flink</groupId>
  <artifactId>flink-clients_2.11</artifactId>
  <version>1.13.6</version>
</dependency>

二进制文件（Binaries）

您可以在更新过的下载页面上找到这些二进制文件。

Docker 镜像（Docker Images）

library/flink (官方镜像，official images)
apache/flink (ASF 仓库，ASF repository)

PyPi

apache-flink==1.13.6

发行说明（Release Notes）

Bug 修复

[FLINK-15987] - SELECT 1.0e0 / 0.0e0 抛出 NumberFormatException 异常
[FLINK-17914] - HistoryServer deletes cached archives if archive listing fails
[FLINK-20195] - Jobs endpoint returns duplicated jobs
[FLINK-20370] - Result is wrong when sink primary key is not the same with query
[FLINK-21289] - Application mode ignores the pipeline.classpaths configuration
[FLINK-23919] - PullUpWindowTableFunctionIntoWindowAggregateRule generates invalid Calc for Window TVF
[FLINK-24232] - Archiving of suspended jobs prevents breaks subsequent archive attempts
[FLINK-24255] - Test Environment / Mini Cluster do not forward configuration.
[FLINK-24310] - A bug in the BufferingSink example in the doc
[FLINK-24318] - Casting a number to boolean has different results between ‘select’ fields and ‘where’ condition
[FLINK-24334] - Configuration kubernetes.flink.log.dir not working
[FLINK-24366] - Unnecessary/misleading error message about failing restores when tasks are already canceled.
[FLINK-24401] - Metaspace 发生 OOM 后，TM 无法退出
[FLINK-24465] - Wrong javadoc and documentation for buffer timeout
[FLINK-24492] - incorrect implicit type conversion between numeric and (var)char
[FLINK-24506] - checkpoint directory is not configurable through the Flink configuration passed into the StreamExecutionEnvironment
[FLINK-24509] - FlinkKafkaProducer example is not compiling due to incorrect constructer signature used
[FLINK-24540] - Fix Resource leak due to Files.list
[FLINK-24543] - Zookeeper connection issue causes inconsistent state in Flink
[FLINK-24563] - Comparing timstamp_ltz with random string throws NullPointerException
[FLINK-24597] - RocksdbStateBackend getKeysAndNamespaces would return duplicate data when using MapState
[FLINK-24621] - JobManager fails to recover 1.13.1 checkpoint due to InflightDataRescalingDescriptor
[FLINK-24662] - PyFlink sphinx check failed with “node class ‘meta’ is already registered, its visitors will be overridden”
[FLINK-24667] - Channel state writer would fail the task directly if meeting exception previously
[FLINK-24676] - Schema does not match if explain insert statement with partial column
[FLINK-24678] - Correct the metric name of map state contains latency
[FLINK-24708] - ConvertToNotInOrInRule has a bug which leads to wrong result
[FLINK-24728] - Batch SQL file sink forgets to close the output stream
[FLINK-24761] - Fix PartitionPruner code gen compile fail
[FLINK-24846] - AsyncWaitOperator fails during stop-with-savepoint
[FLINK-24860] - Fix the wrong position mappings in the Python UDTF
[FLINK-24885] - ProcessElement Interface parameter Collector : java.lang.NullPointerException
[FLINK-24922] - Fix spelling errors in the word “parallism”
[FLINK-25022] - ClassLoader leak with ThreadLocals on the JM when submitting a job through the REST API
[FLINK-25067] - Correct the description of RocksDB’s background threads
[FLINK-25084] - Field names must be unique. Found duplicates
[FLINK-25091] - Official website document FileSink orc compression attribute reference error
[FLINK-25096] - Issue in exceptions API(/jobs/:jobid/exceptions) in flink 1.13.2
[FLINK-25199] - StreamEdges are not unique in self-union, which blocks propagation of watermarks
[FLINK-25362] - Incorrect dependencies in Table Confluent/Avro docs
[FLINK-25468] - Local recovery fails if local state storage and RocksDB working directory are not on the same volume
[FLINK-25486] - Perjob can not recover from checkpoint when zookeeper leader changes
[FLINK-25494] - Duplicate element serializer during DefaultOperatorStateBackendSnapshotStrategy#syncPrepareResources
[FLINK-25513] - CoFlatMapFunction requires both two flat_maps to yield something
[FLINK-25559] - SQL JOIN causes data loss
[FLINK-25683] - wrong result if table transfrom to DataStream then window process in batch mode
[FLINK-25728] - Potential memory leaks in StreamMultipleInputProcessor
[FLINK-25732] - Dispatcher#requestMultipleJobDetails returns non-serialiable collection

优化（Improvement）

[FLINK-21407] - 明确哪些源和 API 支持哪些格式
[FLINK-20443] - ContinuousProcessingTimeTrigger 在窗口结束时不会触发
[FLINK-21467] - Document possible recommended usage of Bounded{One/Multi}Input.endInput and emphasize that they could be called multiple times
[FLINK-23842] - Add log messages for reader registrations and split requests.
[FLINK-24631] - Avoiding directly use the labels as selector for deployment and service
[FLINK-24739] - State requirements for Flink’s application mode in the documentation
[FLINK-24987] - Enhance ExternalizedCheckpointCleanup enum
[FLINK-25160] - Make doc clear: tolerable-failed-checkpoints counts consecutive failures
[FLINK-25415] - implement retrial on connections to Cassandra container
[FLINK-25611] - Remove CoordinatorExecutorThreadFactory thread creation guards
[FLINK-25818] - Add explanation how Kafka Source deals with idleness when parallelism is higher then the number of partitions