Apache Spark 1.6.2 发布了，Apache Spark 是一种与 Hadoop 相似的开源集群计算环境，但是两者之间还存在一些不同之处，这些有用的不同之处使 Spark 在某些工作负载方面表现得更加优越，换句话说，Spark 启用了内存分布数据集，除了能够提供交互式查询外，它还可以优化迭代工作负载。
Spark 是在 Scala 语言中实现的，它将 Scala 用作其应用程序框架。与 Hadoop 不同，Spark 和 Scala 能够紧密集成，其中的 Scala 可以像操作本地集合对象一样轻松地操作分布式数据集。
尽管创建 Spark 是为了支持分布式数据集上的迭代作业，但是实际上它是对 Hadoop 的补充，可以在 Hadoo 文件系统中并行运行。
改进日志如下：
Sub-task
[SPARK-15613] - Incorrect days to millis conversion
[SPARK-15723] - SimpleDateParamSuite test is locale-fragile and relies on deprecated short TZ name
Bug
[SPARK-8428] - TimSort Comparison method violates its general contract with CLUSTER BY
[SPARK-10722] - Uncaught exception: RDDBlockId not found in driver-heartbeater
[SPARK-11327] - spark-dispatcher doesn't pass along some spark properties
[SPARK-11507] - Error thrown when using BlockMatrix.add
[SPARK-12655] - GraphX does not unpersist RDDs
[SPARK-12712] - test-dependencies.sh script fails when run against empty .m2 cache
[SPARK-12941] - Spark-SQL JDBC Oracle dialect fails to map string datatypes to Oracle VARCHAR datatype
[SPARK-13023] - Check for presence of 'root' module after computing test_modules, not changed_modules
[SPARK-13207] - _SUCCESS should not break partition discovery
[SPARK-13227] - Risky apply() in OpenHashMap
[SPARK-13242] - Moderately complex `when` expression causes code generation failure
[SPARK-13327] - colnames()<- allows invalid column names
[SPARK-13352] - BlockFetch does not scale well on large block
[SPARK-13444] - QuantileDiscretizer chooses bad splits on large DataFrames
[SPARK-13522] - Executor should kill itself when it's unable to heartbeat to the driver more than N times
[SPARK-13566] - Deadlock between MemoryStore and BlockManager
[SPARK-13622] - Issue creating level db file for YARN shuffle service if URI is used in yarn.nodemanager.local-dirs
[SPARK-13631] - getPreferredLocations race condition in spark 1.6.0?
[SPARK-13642] - Properly handle signal kill of ApplicationMaster
[SPARK-13648] - org.apache.spark.sql.hive.client.VersionsSuite fails NoClassDefFoundError on IBM JDK
[SPARK-13652] - TransportClient.sendRpcSync returns wrong results
[SPARK-13697] - TransformFunctionSerializer.loads doesn't restore the function's module name if it's '__main__'
[SPARK-13705] - UpdateStateByKey Operation documentation incorrectly refers to StatefulNetworkWordCount
[SPARK-13711] - Apache Spark driver stopping JVM when master not available
[SPARK-13755] - Escape quotes in SQL plan visualization node labels
[SPARK-13772] - DataType mismatch about decimal
[SPARK-13803] - Standalone master does not balance cluster-mode drivers across workers
[SPARK-13806] - SQL round() produces incorrect results for negative values
[SPARK-13845] - BlockStatus and StreamBlockId keep on growing result driver OOM
[SPARK-13850] - TimSort Comparison method violates its general contract
[SPARK-13901] - We get wrong logdebug information when jump to the next locality level.
[SPARK-13958] - Executor OOM due to unbounded growth of pointer array in Sorter
[SPARK-14006] - Builds of 1.6 branch fail R check
[SPARK-14074] - Use fixed version of install_github in SparkR build
[SPARK-14159] - StringIndexerModel sets output column metadata incorrectly
[SPARK-14187] - Incorrect use of binarysearch in SparseMatrix
[SPARK-14204] - [SQL] Failure to register URL-derived JDBC driver on executors in cluster mode
[SPARK-14219] - Fix `pickRandomVertex` not to fall into infinite loops for graphs with one vertex
[SPARK-14232] - Event timeline on job page doesn't show if an executor is removed with multiple line reason
[SPARK-14243] - updatedBlockStatuses does not update correctly when removing blocks
[SPARK-14261] - Memory leak in Spark Thrift Server
[SPARK-14298] - LDA should support disable checkpoint
[SPARK-14322] - Use treeAggregate instead of reduce in OnlineLDAOptimizer
[SPARK-14357] - Tasks that fail due to CommitDeniedException (a side-effect of speculation) can cause job failure
[SPARK-14363] - Executor OOM due to a memory leak in Sorter
[SPARK-14368] - Support python.spark.worker.memory with upper-case unit
[SPARK-14454] - Better exception handling while marking tasks as failed
[SPARK-14468] - Always enable OutputCommitCoordinator
[SPARK-14495] - Distinct aggregation cannot be used in the having clause
[SPARK-14563] - SQLTransformer.transformSchema is not implemented correctly
[SPARK-14665] - PySpark StopWordsRemover default stopwords are Java object
[SPARK-14671] - Pipeline.setStages needs to handle Array non-covariance
[SPARK-14679] - UI DAG visualization causes OOM generating data
[SPARK-14739] - Vectors.parse doesn't handle dense vectors of size 0 and sparse vectors with no indices
[SPARK-14757] - Incorrect behavior of Join operation in Spqrk SQL JOIN : "false" in the left table is joined to "null" on the right table
[SPARK-14915] - Tasks that fail due to CommitDeniedException (a side-effect of speculation) can cause job to never complete
[SPARK-14965] - StructType throws exception for missing field
[SPARK-15165] - Codegen can break because toCommentSafeString is not actually safe
[SPARK-15209] - Web UI's timeline visualizations fails to render if descriptions contain single quotes
[SPARK-15260] - UnifiedMemoryManager could be in bad state if any exception happen while evicting blocks
[SPARK-15262] - race condition in killing an executor and reregistering an executor
[SPARK-15528] - conv function returns inconsistent result for the same data
[SPARK-15601] - CircularBuffer's toString() to print only the contents written if buffer isn't full
[SPARK-15736] - Gracefully handle loss of DiskStore files
[SPARK-15754] - org.apache.spark.deploy.yarn.Client changes the credential of current user
[SPARK-15892] - Incorrectly merged AFTAggregator with zero total count
[SPARK-15975] - Improper Popen.wait() return code handling in dev/run-tests
[SPARK-16017] - YarnClientSchedulerBackend now registers backends as IPs instead of Hostnames which causes all tasks to run with RACK_LOCAL locality.
[SPARK-16035] - The SparseVector parser fails checking for valid end parenthesis
[SPARK-16086] - Python UDF failed when there is no arguments
[SPARK-16173] - Can't join describe() of DataFrame in Scala 2.10
Documentation
[SPARK-14618] - RegressionEvaluator doc out of date
[SPARK-15223] - spark.executor.logs.rolling.maxSize wrongly referred to as spark.executor.logs.rolling.size.maxBytes
Improvement
[SPARK-13599] - Groovy-all ends up in spark-assembly if hive profile set
[SPARK-13601] - Invoke task failure callbacks before calling outputstream.close()
[SPARK-13663] - Upgrade Snappy Java to 1.1.2.1
[SPARK-13810] - Add Port Configuration Suggestions on Bind Exceptions
[SPARK-14058] - Incorrect docstring in Window.orderBy
[SPARK-14107] - PySpark spark.ml GBT algs need seed Param
[SPARK-14149] - Log exceptions in tryOrIOException
[SPARK-14242] - avoid too many copies in network when a network frame is large
[SPARK-14787] - Upgrade Joda-Time library from 2.9 to 2.9.3
[SPARK-15205] - Codegen can compile the same source code more than twice
[SPARK-15827] - Publish Spark's forked sbt-pom-reader to Maven Central
New Feature
[SPARK-11515] - QuantileDiscretizer should take random seed
[SPARK-13465] - Add a task failure listener to TaskContext
软件详情：http://spark.apache.org/
下载地址：http://spark.apache.org/downloads.html
来自:开源中国社区
