ed18e97 | Tzu-Li (Gordon) Tai | 17 March 2017, 10:54:11 UTC | Commit for release 1.1.5 | 17 March 2017, 10:54:11 UTC |
6662cc6 | Tzu-Li (Gordon) Tai | 15 March 2017, 16:05:51 UTC | [FLINK-5701] [kafka] FlinkKafkaProducer should check asyncException on checkpoints This closes #3549. | 16 March 2017, 04:41:20 UTC |
e296aca | Tzu-Li (Gordon) Tai | 10 March 2017, 06:47:57 UTC | [FLINK-6006] [kafka] Always use complete restored state in FlinkKafkaConsumer Previously, the Kafka Consumer performs partition list querying on restore, and then uses it to filter out restored state of partitions that doesn't exist in the list. If in any case the returned partitions list is incomplete (i.e. missing partitions that existed before perhaps due to temporary ZK / broker downtimes), then the state of the missing partitions is dropped and cannot be recovered anymore. This commit fixes this by always restoring the complete state, without any sort of filtering. We simply let the consumer fail if assigned partitions to the consuming threads / Kafka clients are unreachable when the consumer starts running. This closes #3507. | 15 March 2017, 14:35:22 UTC |
a34559d | Till Rohrmann | 01 March 2017, 13:08:35 UTC | [FLINK-5940] [checkpoint] Harden ZooKeeperCompletedCheckpointStore.recover method The ZooKeeperCompletedCheckpointStore only tries to recover the latest completed checkpoint even though it might have read older checkpoint state handles from ZooKeeper. In order to deal with corrupted state handles, this commit changes the behaviour such that the completed checkpoint store adds all read retrievable state handles from ZooKeeper and upon request of the latest checkpoint it will return the latest completed checkpoint which could be retrieved from the state handles. Broken state handles are removed from the completed checkpoint store and ZooKeeper. This closes #3451. | 14 March 2017, 08:26:15 UTC |
97616fd | Till Rohrmann | 01 March 2017, 17:03:41 UTC | [FLINK-5942] [checkpoint] Harden ZooKeeperStateHandleStore to handle corrupt data If calling ZooKeeperStateHandleStore.getAll or getAllSortedByName as the ZooKeeperCompletedCheckpointStore does in the recovery case, the operation will fail if there exists a Znode with corrupted data. This will break Flink's recovery mechanism, because it will read this node over and over again. In order to solve this problem, this commit changes the behaviour such that it ignores corrupted Znodes it cannot read. This closes #3452. | 14 March 2017, 08:19:47 UTC |
e50bf65 | Tzu-Li (Gordon) Tai | 12 March 2017, 14:46:27 UTC | [FLINK-6025] [core] Add Flink's own JavaSerializer for Kryo serialization This commit adds a reimplemented JavaSerializer to be registered with Kryo. This is due to a know issue with Kryo's JavaSerializer that may use the wrong classloader for deserialzation. Instead of registering Kryo's JavaSerializer for Throwables, it is now changed to register the reimplemented JavaSerializer. Users who bump into ClassNotFoundExceptions if they are using Kryo's JavaSerializer for their own types are also recommended to change to Flink's JavaSerializer. This closes #3519. | 13 March 2017, 15:42:38 UTC |
01703e6 | Greg Hogan | 01 March 2017, 20:55:48 UTC | [FLINK-5945] [core] Close function in OuterJoinOperatorBase#executeOnCollections Conclude OuterJoinOperatorBase#executeOnCollections with a call to FunctionUtils.closeFunction(function) in order to close rich user functions. This closes #3453 | 02 March 2017, 14:42:22 UTC |
ba5aa10 | Till Rohrmann | 28 February 2017, 14:20:47 UTC | [FLINK-5934] Set the Scheduler in the ExecutionGraph via its constructor Before the scheduler was set when calling ExecutionGraph.scheduleForExecution(). This has the disadvantage that the ExecutionGraph has not scheduler set if something else went wrong before the scheduleForExecution call. Consequently, the job will be stuck in a restart loop because the recovery will fail if there is no Scheduler set. In order to solve the problem, the Scheduler is not passed to the ExecutionGraph when it is created. This closes #3441. | 02 March 2017, 13:02:59 UTC |
44f48b3 | Fabian Hueske | 14 February 2017, 21:02:26 UTC | [FLINK-5771] [core] Fix multi-char delimiter detection in DelimitedInputFormat. - Add a test case to validate correct delimiter detection. - Remove a couple of try-catch blocks from existing tests. This closes #3316. | 15 February 2017, 22:59:46 UTC |
3cd7c8e | David Anderson | 31 January 2017, 14:15:15 UTC | [FLINK-5575] [docs] Add outdated release warning | 06 February 2017, 16:23:35 UTC |
3f2860f | Ufuk Celebi | 03 February 2017, 09:23:23 UTC | [docs] Set version to 1.1.4 | 03 February 2017, 09:23:23 UTC |
e1861db | Aljoscha Krettek | 24 January 2017, 15:18:42 UTC | [FLINK-5647] Fix RocksDB Backend Cleanup | 25 January 2017, 22:49:16 UTC |
0c99e48 | Stephan Ewen | 25 January 2017, 11:15:49 UTC | [FLINK-5639] [rabbitmq connector] Fix incorrect location of README.md for backported notice | 25 January 2017, 11:15:49 UTC |
743aaf6 | Stephan Ewen | 25 January 2017, 11:06:12 UTC | [FLINK-5639] [rabbitmq connector] Add a note about MPL 1.1 license of Maven dependency | 25 January 2017, 11:12:08 UTC |
6566b63 | Fabian Hueske | 05 January 2017, 23:00:30 UTC | [FLINK-2662] [optimizer] Fix translation of broadcasted unions. | 20 January 2017, 16:35:44 UTC |
f6f1c24 | Stephan Ewen | 20 January 2017, 10:12:12 UTC | [FLINK-5585] [jobmanager] Fix NullPointerException in JobManager.updateAccumulators | 20 January 2017, 10:14:35 UTC |
931929b | Ufuk Celebi | 17 January 2017, 18:10:33 UTC | [FLINK-5484] [serialization] Add test for registered Kryo types | 18 January 2017, 14:28:10 UTC |
214c188 | Jakub Havlik | 17 January 2017, 07:26:07 UTC | [FLINK-5518] [hadoopCompat] Add null check to HadoopInputFormatBase.close(). | 17 January 2017, 19:59:26 UTC |
4ea52d6 | Ufuk Celebi | 12 January 2017, 14:54:40 UTC | [FLINK-5466] [webfrontend] Rebuild CSS/JS files | 12 January 2017, 14:54:40 UTC |
12cf5dc | Ufuk Celebi | 12 January 2017, 14:47:51 UTC | [FLINK-5466] [webfrontend] Set environment to production in gulpfile | 12 January 2017, 14:53:31 UTC |
a874ad8 | Ufuk Celebi | 16 December 2016, 13:49:07 UTC | [FLINK-5352] [rocksdb] Restore 1.1.3 RocksDB DBOptions | 16 December 2016, 13:51:09 UTC |
d31c9fa | kl0u | 14 December 2016, 15:17:37 UTC | [FLINK-5292] Add savepointToFile in OneInputStreamOperatorTestHarness. This will be used mainly for migration testing purposes between Flink 1.1 and Flink 1.2. | 16 December 2016, 13:16:22 UTC |
a8b415f | Stephan Ewen | 16 December 2016, 11:32:02 UTC | [FLINK-5352] [rocksdb] Restore 1.1.3 RocksDB memory footprint | 16 December 2016, 13:07:43 UTC |
9ed7752 | Ufuk Celebi | 13 December 2016, 10:26:47 UTC | [FLINK-5326] [network] Log errors in sending Netty handler | 13 December 2016, 13:01:20 UTC |
04db15a | Ufuk Celebi | 13 December 2016, 10:26:47 UTC | [FLINK-5326] [network] Check release flag of parent in reader In PipelinedSubpartitionView, there is a possible race with releasing the parent subpartition and querying for a buffer in the view. The parent partition release clears all buffers in locked scope and releases the view outside of the lock. If concurrently the view is queried for a buffer it might get null, which is only allowed if the view was released. Because the release is only forwarded out of the lock scope, this can happen before the release has propagated. As a solution, we check the parent release status as well in the view. This is how it is handled in the spilled views, too. This surfaced with the recent refactorings, because the previous consumption model required multiple rounds of get, registerListener, isReleased calls, which hid this problem. | 13 December 2016, 12:59:47 UTC |
c0d5f1d | Maximilian Michels | 21 November 2016, 11:46:51 UTC | [FLINK-5071][yarn] adjust vcore validation check The check didn't take the virtual core settings configured in the Flink configuration into account. - improve error reporting - add test case This closes #2839. | 12 December 2016, 13:46:02 UTC |
f3d0cc3 | Till Rohrmann | 08 December 2016, 17:53:40 UTC | [FLINK-5300] Add more gentle file deletion procedure Before deleting a parent directory always check the directory whether it contains some files. If not, then try to delete the parent directory. This will give a more gentle behaviour wrt storage systems which are not instructed to delete a non-empty directory. This closes #2971. | 12 December 2016, 11:06:13 UTC |
afaa27e | Till Rohrmann | 07 December 2016, 18:05:47 UTC | [FLINK-5285] Abort checkpoint only once in BarrierTracker Prevent an interleaved sequence of cancellation markers for two consecutive checkpoints to trigger a flood of cancellation markers for down stream operators. This is done by aborting each checkpoint only once and don't re-create checkpoint barrier counts for already aborted checkpoints. Add test case This closes #2964. | 09 December 2016, 13:37:31 UTC |
2b612f2 | Ufuk Celebi | 08 December 2016, 22:48:39 UTC | [FLINK-5114] [network] Handle partition producer state check for unregistered executions This closes #2975. | 09 December 2016, 13:20:08 UTC |
3ae6e9e | Robert Metzger | 06 December 2016, 20:03:10 UTC | [FLINK-5039] Bump Avro version to 1.7.7. This closes #2953. | 08 December 2016, 20:24:15 UTC |
671b434 | Boris Osipov | 08 December 2016, 13:58:21 UTC | [FLINK-5169] [network] Fix String formats in spillable partition This closes #2967. | 08 December 2016, 14:22:08 UTC |
b046038 | Till Rohrmann | 07 December 2016, 15:22:23 UTC | [FLINK-5278] Improve task and checkpoint related logging This closes #2690. | 07 December 2016, 17:07:55 UTC |
33df945 | Till Rohrmann | 07 December 2016, 16:28:20 UTC | [FLINK-4631] Prevent NPE in TwoInputStreamTask Check that the input processor has been created before cleaning it up. | 07 December 2016, 16:29:58 UTC |
4526005 | Ufuk Celebi | 07 December 2016, 12:48:25 UTC | [FLINK-5275] [execgraph] Give more detailed error message if InputChannel deployment fails | 07 December 2016, 16:13:01 UTC |
1b472d2 | Ufuk Celebi | 07 December 2016, 12:39:18 UTC | [FLINK-5274] [network] Handle reader release in LocalInputChannel | 07 December 2016, 16:13:01 UTC |
75b48ed | Ufuk Celebi | 07 December 2016, 13:26:35 UTC | [FLINK-5276] [eg] Check for null when archiving prior execution attempts This closes #2957. | 07 December 2016, 16:03:19 UTC |
f46b7a0 | zentol | 05 December 2016, 14:22:22 UTC | [hotfix] [metris] [docs] Correct scope format keys This closes #2947. | 06 December 2016, 13:04:18 UTC |
7bb6b69 | David Anderson | 05 December 2016, 10:39:46 UTC | [FLINK-5259] [docs] Fix wrong execution environment in batch retry delays example. This closes #2943. | 06 December 2016, 12:45:14 UTC |
c567b0c | David Anderson | 04 December 2016, 14:32:23 UTC | [FLINK-5249] [docs] Fix description of datastream rescaling to match the figure. This closes #2932. | 06 December 2016, 12:45:03 UTC |
c7db532 | Maximilian Michels | 05 December 2016, 13:32:11 UTC | [FLINK-5262][docs] Introduce Gemfile.lock to avoid dependency range conflicts The Gemfile for specifying the Ruby dependencies of our documentation has fixed dependency versions to avoid incompatible changes with different versions of the dependencies. However, Ruby's dependency management allows artifacts to specify ranges for dependencies. This can be problematic. For instance, we use 'jekyll' version 2.5.3 which depends on 'jekyll-gist' ~> 1.0 which means 1.0 >= version < 2.0. This may resolve 'jekyll-gist' 1.4.0 which depends on 'octokit' ~> 4.2 which may be 4.2 >= versions < 5.0. Too bad, 'octokit' starting with 4.4 depends on Ruby version >= 2.0 which is not available on our build servers. Since we already use the improved version of Rubys build system called 'bundler', we can mitigate this problem by checking in a Gemfile.lock file which specifies the exact versions of all dependencies required to build the docs. This closes #2945. | 05 December 2016, 17:07:25 UTC |
a5065e3 | Ufuk Celebi | 05 December 2016, 10:50:38 UTC | [FLINK-5248] [tests] Catch restore failures in SavepointITCase - Minor test clean up and reduced checkpointing interval for speed up - The test did not catch a task restore failure since only the TDDs were tested. Now, we test that restore is actually called and some checkpoints complete after restoring from a savepoint. | 05 December 2016, 10:50:39 UTC |
e4ca3a5 | Rohit Agarwal | 03 December 2016, 04:21:38 UTC | [FLINK-2844] [docs] Remove obsolete 'new-web-frontend' option This closes #2927. | 04 December 2016, 14:47:36 UTC |
951a3b6 | Rohit Agarwal | 02 December 2016, 23:01:00 UTC | [docs] Replace broken download link This variable was removed in dc5062557a55 | 04 December 2016, 14:47:13 UTC |
da09d41 | Till Rohrmann | 03 December 2016, 19:15:35 UTC | [FLINK-5246] Don't discard checkpoint messages if they are unknown This is the case if the savepoint coordinator has triggered a checkpoint. The corresponding checkpoint messages are not known to the checkpoint coordinator and thus should not be discarded. Instead, the JobManager will now discard all messages which have not been accepted by neither the CheckpointCoordinator nor the SavepointCoordinator. This closes #2930. | 04 December 2016, 11:38:32 UTC |
020da2c | Till Rohrmann | 02 December 2016, 14:33:06 UTC | [FLINK-5229] [tm] Discard completed checkpoints if a subsequent operator's checkpoint fails In case of a failure of any StreamOperator#snapshotState method, all up to this point created StreamTaskStates are discarded. This ensures that a failing checkpoint operation of a chained operator won't leave orphaned checkpoint data behind. This closes #2924. | 02 December 2016, 18:10:46 UTC |
388acbc | Ufuk Celebi | 02 December 2016, 11:52:58 UTC | [FLINK-5228] [network] Fix LocalInputChannel re-trigger request and release deadlock | 02 December 2016, 14:17:55 UTC |
0dc82ba | Till Rohrmann | 02 December 2016, 08:46:00 UTC | [hotfix] Add null check to CheckpointCoordinator#discardState | 02 December 2016, 09:37:36 UTC |
2bf8722 | Ufuk Celebi | 01 December 2016, 17:38:30 UTC | [FLINK-5169] [network] Fix spillable subpartition buffer count | 01 December 2016, 20:41:55 UTC |
9c05887 | Till Rohrmann | 01 December 2016, 12:25:05 UTC | [FLINK-5214] Clean up checkpoint data in case of a failing checkpoint operation Adds exception handling to the stream operators for the snapshotState method. In case of an exception while performing the snapshot operation, all until then checkpointed data will be discarded/deleted. This makes sure that a failing checkpoint operation won't leave orphaned checkpoint data (e.g. files) behind. Add test case for FsCheckpointStateOutputStream Add RocksDB FullyAsyncSnapshot cleanup test Add proper state cleanup tests for window operator Add state cleanup test for failing snapshot call of AbstractUdfStreamOperator | 01 December 2016, 17:04:43 UTC |
4b734d7 | Till Rohrmann | 24 November 2016, 17:16:28 UTC | [FLINK-5158] [ckPtCoord] Handle exceptions from CompletedCheckpointStore in CheckpointCoordinator Handle exceptions from the CompletedCheckpointStore properly in the CheckpointCoordinator. This means that in case of an exception, the completed checkpoint will be properly cleaned up and also the triggering of subsequent checkpoints will be started. Fix failing SavepointCoordinatorTest | 01 December 2016, 17:00:53 UTC |
d314bc5 | Till Rohrmann | 29 November 2016, 16:31:08 UTC | [FLINK-5193] [jm] Harden job recovery in case of recovery failures When recovering multiple jobs a single recovery failure caused all jobs to be not recovered. This PR changes this behaviour to make the recovery of jobs independent so that a single failure won't stall the complete recovery. Furthermore, this PR improves the error reporting for failures originating in the ZooKeeperSubmittedJobGraphStore. Add test case Fix failing JobManagerHACheckpointRecoveryITCase | 01 December 2016, 16:53:34 UTC |
59f61bf | Stephan Ewen | 30 November 2016, 21:38:23 UTC | [FLINK-5218] [state backends] Eagerly close pending FsCheckpointStateOutputStream on task cancellation This fix contains two modifications: 1. State backends implement 'Closeable' and register themselves at the 'canceleables' 2. The FsStateBackend tracks all it unclosed FsCheckpointOutputStreams and closes them on 'close()' | 01 December 2016, 14:09:43 UTC |
e475eb2 | Stephan Ewen | 30 November 2016, 19:31:07 UTC | [FLINK-5216] [checkpoints] 'Min Time Between Checkpoints' references timestamp after checkpoint | 01 December 2016, 13:01:40 UTC |
7b5d769 | Ivan Mushketyk | 30 November 2016, 20:31:39 UTC | [FLINK-3680] [web frontend] Remove "(not set)" text in the Job Plan UI This closes #2457 | 30 November 2016, 20:31:39 UTC |
7de7bfd | Greg Hogan | 30 November 2016, 20:30:12 UTC | [hotfix] [webfrontend] Rebuild | 30 November 2016, 20:30:12 UTC |
357b932 | Ufuk Celebi | 29 November 2016, 16:14:23 UTC | [FLINK-5192] [logging] Improve log config templates This closes #2899. | 30 November 2016, 15:22:11 UTC |
28e44e7 | Ufuk Celebi | 29 November 2016, 16:08:53 UTC | [FLINK-5207] [logging] Decrease HadoopFileSystem logging | 30 November 2016, 15:22:11 UTC |
8d949c9 | Ufuk Celebi | 29 November 2016, 15:35:14 UTC | [FLINK-5199] [logging] Improve logging in ZooKeeperSubmittedJobGraphStore | 30 November 2016, 15:22:11 UTC |
ee478fe | Ufuk Celebi | 29 November 2016, 15:15:30 UTC | [FLINK-5198] [logging] Improve TaskState toString | 30 November 2016, 15:22:11 UTC |
7b9a444 | Ufuk Celebi | 29 November 2016, 15:04:48 UTC | [FLINK-5196] [logging] Don't log InputChannelDeploymentDescriptor | 30 November 2016, 15:22:11 UTC |
c8ade63 | Ufuk Celebi | 29 November 2016, 15:00:02 UTC | [FLINK-5201] [logging] Log loaded config properties on INFO level | 30 November 2016, 15:22:11 UTC |
ffe6b6b | Ufuk Celebi | 29 November 2016, 16:15:27 UTC | [FLINK-5194] [logging] Log heartbeats on TRACE level | 30 November 2016, 15:22:11 UTC |
569a966 | Till Rohrmann | 29 November 2016, 15:02:29 UTC | [FLINK-5197] [jm] Ignore outdated JobStatusChanged messages Outdated JobStatusChanged messages no longer trigger a RemoveJob message but are logged and ignored. This has the advantage, that an outdated JobStatusChanged message cannot interfere with a recovered job which can have the same job id. This closes #2896. | 30 November 2016, 12:41:05 UTC |
0758d0b | godfreyhe | 29 November 2016, 11:27:58 UTC | [FLINK-5184] [table] Fix compareSerialized() of RowComparator. | 29 November 2016, 14:43:57 UTC |
5ebd7c8 | Stephan Ewen | 27 November 2016, 17:15:40 UTC | [FLINK-5169] [network] Add tests for channel consumption This closes #2882. | 28 November 2016, 20:05:00 UTC |
8d97eaa | Ufuk Celebi | 28 November 2016, 08:59:58 UTC | [FLINK-5169] [network] Adjust tests to new consumer logic | 28 November 2016, 20:05:00 UTC |
6cfce17 | Stephan Ewen | 28 November 2016, 08:59:29 UTC | [FLINK-5169] [network] Make consumption of InputChannels fair | 28 November 2016, 20:05:00 UTC |
e2c53cf | Stephan Ewen | 28 November 2016, 14:13:35 UTC | [FLINK-5142] [checkpoints] Fix timer resource leak in CheckpointCoordinator | 28 November 2016, 14:13:35 UTC |
ca90809 | sergey_sokur | 17 November 2016, 16:28:20 UTC | [FLINK-5050] [build] Remove transitive JSON.org dependency This transitive dependency has an incompatible license. This closes #2824 | 28 November 2016, 11:09:24 UTC |
6377ef0 | Stephan Ewen | 28 November 2016, 10:38:37 UTC | [FLINK-5173] [RocksDB backend] Upgrade RocksDB version to 4.11.2 This fixes some bugs in RocksDB that may cause data corruption in some cases (observed for very long values) and also adds support for Microsoft Windows. | 28 November 2016, 10:38:37 UTC |
9410d57 | Stephan Ewen | 28 November 2016, 10:21:45 UTC | [FLINK-5172] [RockDB backend] Set dependencies to 'flink-core' and 'flink-streaming-java' as 'provided' | 28 November 2016, 10:21:45 UTC |
3b5d3c6 | Ufuk Celebi | 12 November 2016, 19:49:17 UTC | Revert "[FLINK-4913][yarn] include user jars in system class loader" This reverts commit ea41b9c56fdc0af3c97d6dd48d04218db6176ec8. This closes #2795 | 25 November 2016, 15:14:52 UTC |
b9e6dcc | Tzu-Li (Gordon) Tai | 17 November 2016, 06:24:24 UTC | [FLINK-5075] [kinesis] Make connector fail-proof to incorrect Kinesalite API behaviour This closes #2822. | 24 November 2016, 10:54:07 UTC |
fc24c30 | Maximilian Michels | 23 November 2016, 15:01:35 UTC | [FLINK-5149] let ContinuousEventTimeTrigger fire at the end of the window This changes the ContinuousEventTimeTrigger to behave like the EventTimeTrigger in the sense that it also triggers at the end of the window. This prevents the trigger from not firing at all in case the first trigger interval is after the window end. This closes #2860. [typo] fix toString() of ContinuousEventTimeTrigger This closes #2854. | 24 November 2016, 10:38:42 UTC |
424fb24 | Robert Metzger | 04 November 2016, 15:15:26 UTC | [hotfix] [streaming] Fix type extraction for joined streams. | 23 November 2016, 20:08:58 UTC |
074b4e6 | Robert Metzger | 04 November 2016, 14:41:06 UTC | [hotfix][docs] Stream joins don't support tuple position keys | 23 November 2016, 20:08:29 UTC |
efbd293 | Fabian Hueske | 21 November 2016, 18:06:42 UTC | [FLINK-2662] [optimizer] Fix computation of global properties of union operator. - Fixes invalid shipping strategy between consecutive unions. | 23 November 2016, 20:08:08 UTC |
871de0b | Till Rohrmann | 22 November 2016, 23:19:32 UTC | [FLINK-5107] Handle evicted execution attempts in request handlers If a prior execution attempt cannot be retrieved because it has been evicted before, the request handler will now throw a meaningful exception to notify the requester about the evicted execution attempt. | 22 November 2016, 23:46:40 UTC |
8989a9f | Stefan Richter | 18 November 2016, 18:07:56 UTC | [FLINK-5107] Introduced limit for prior execution attempt history This closes #2837. | 22 November 2016, 23:29:20 UTC |
a79290c | Stephan Ewen | 22 November 2016, 11:54:17 UTC | [FLINK-5123] [docs] Add description how to build a properly shaded build with different Maven versions | 22 November 2016, 11:56:22 UTC |
cf4b221 | Till Rohrmann | 17 November 2016, 14:39:11 UTC | [FLINK-5085] Execute CheckpointCoordinator's state discard calls asynchronously The CheckpointCoordinator is now given an Executor which is used to execute the state discard calls asynchronously. This will prevent blocking operations to be executed from within the calling thread. Shut down ExecutorServices gracefully | 22 November 2016, 08:48:00 UTC |
f2e4c19 | Till Rohrmann | 15 November 2016, 21:45:04 UTC | [FLINK-5073] Use Executor to run ZooKeeper callbacks in ZooKeeperStateHandleStore Use dedicated Executor to run ZooKeeper callbacks in ZooKeeperStateHandleStore instead of running it in the ZooKeeper client's thread. The callback can be blocking because it discards state which might entail deleting files from disk. Add TestExecutors Introduce dedicated Executor for blocking io operations | 22 November 2016, 08:47:57 UTC |
7fb71c5 | Till Rohrmann | 16 November 2016, 17:33:54 UTC | [FLINK-5082] Pull ExecutorService lifecycle management out of the JobManager The provided ExecutorService will no longer be closed by the JobManager. Instead the lifecycle is managed outside of it where it was created. This will give a nicer behaviour, because it better seperates responsibilities. | 21 November 2016, 14:53:19 UTC |
68585d1 | iteblog | 14 November 2016, 11:04:28 UTC | [FLINK-5058] [scala-shell] Set correct value for TaskManager memory parameter. This closes #2799. | 17 November 2016, 22:47:19 UTC |
3d31311 | Rohit Agarwal | 17 November 2016, 00:49:57 UTC | [hotfix] [docs] Fix incorrect URL. This closes #2821. | 17 November 2016, 22:47:11 UTC |
bd71b63 | Stephan Ewen | 17 November 2016, 16:34:09 UTC | [hotfix] [checkpoints] Enhance debug logging in exactly-once and at-least-once checkpoint stream aligners | 17 November 2016, 16:42:45 UTC |
4daf3bb | Till Rohrmann | 14 November 2016, 17:33:55 UTC | [FLINK-5063] [checkpointing] Discard state handles of declined or expired state handles Whenever the checkpoint coordinator receives an acknowledge checkpoint message which belongs to the job maintained by the checkpoint coordinator, it should either record the state handles for later processing or discard to free the resources. The latter case can happen if a checkpoint has been expired and late acknowledge checkpoint messages arrive. Furthremore, it can happen if a Task sent a decline checkpoint message while other Tasks where still drawing a checkpoint. This PR changes the behaviour such that state handles belonging to the job of the checkpoint coordinator are discarded if they could not be added to the PendingCheckpoint. This closes #2813 | 16 November 2016, 20:29:32 UTC |
ad3e674 | Ufuk Celebi | 12 November 2016, 19:19:15 UTC | [FLINK-5057] [taskmanager] Read cancellation timeout from task manager config This closes #2794 | 16 November 2016, 20:26:30 UTC |
723ce72 | Tzu-Li (Gordon) Tai | 11 November 2016, 09:01:11 UTC | [FLINK-5013] [kinesis] Shade AWS dependencies to work with older EMR versions This closes #2787. | 16 November 2016, 06:28:37 UTC |
3df5f6c | Ufuk Celebi | 10 November 2016, 21:19:39 UTC | [hotfix] Fix ExecutionVertexDeploymentTest compilation Fixes a compilation error introduced after rebasing the previous commits. | 10 November 2016, 21:19:40 UTC |
0bd8e02 | Ufuk Celebi | 10 November 2016, 10:15:47 UTC | [FLINK-5040] [taskmanager] Adjust partition request backoffs The back offs were hard coded before, which would have made it impossible to react to any potential problems with them. | 10 November 2016, 20:53:31 UTC |
55c506f | Ufuk Celebi | 09 November 2016, 17:25:06 UTC | [FLINK-5040] [jobmanager] Set correct input channel types with eager scheduling This closes #2784. | 10 November 2016, 20:53:30 UTC |
b5a4cb6 | Ufuk Celebi | 10 November 2016, 13:01:22 UTC | Revert "[FLINK-3232] [runtime] Add option to eagerly deploy channels" The reverted commit did not really fix anything, but hid the problem by brute force, sending many more schedule or update consumers messages. | 10 November 2016, 20:53:30 UTC |
9a19ca1 | Till Rohrmann | 09 November 2016, 18:11:36 UTC | [FLINK-5046] [tdd] Preserialize TaskDeploymentDescriptor information In order to speed up the serialization of the TaskDeploymentDescriptor we can pre serialize all information which stays the same for all TaskDeploymentDescriptors. The information which is static for a TDD is the job related information contained in the ExecutionGraph and the operator/task related information stored in the ExecutionJobVertex. In order to pre serialize this information, this PR introduces the JobInformation class and the TaskInformration class which are stored in serialized form in the ExecutionGraph and the ExecutionJobVertex, respectively. Fix for release-1.1 This closes #2780. | 10 November 2016, 18:33:43 UTC |
3eb5633 | Till Rohrmann | 08 November 2016, 14:14:36 UTC | [FLINK-5033] [cep] Advance time with incoming watermarks at CEP operator Before the time was only advanced if the CEP had some events buffered. If the priority queue was empty, then an incoming watermark did not advance the time. This led to missing timeouts and pruning possibilities. The PR fixes this problem. This closes #2771. | 10 November 2016, 13:28:30 UTC |
290f8a2 | Stephan Ewen | 09 November 2016, 12:09:37 UTC | [FLINK-5038] [streaming runtime] Make sure Canceleables are canceled even them "cancelTask" throws an exception | 09 November 2016, 19:24:10 UTC |
32f7efc | Stephan Ewen | 09 November 2016, 15:52:27 UTC | [hotfix] [web frontend] Reduce log level to DEBUG for requests againt unavailable jobs This reduces log noise by a lot in cases where browsers are kept open and re-request the status of old jobs. | 09 November 2016, 19:24:10 UTC |
2041ba0 | Till Rohrmann | 09 November 2016, 11:19:10 UTC | [hotfix] Move logging of JobStatus changes into the ExecutionGraph Prior the JobManager was responsible for logging the JobStatus changes. This introduced out of order logging since the JM was a mere job status listener which was notified by an asynchronous message. | 09 November 2016, 13:57:34 UTC |
0962cb6 | Stephan Ewen | 03 November 2016, 14:28:15 UTC | [FLINK-4975] [checkpointing] Add a limit for how much data may be buffered in alignment. If more data than the defined amount is buffered, the alignment is aborted and the checkpoint canceled. | 08 November 2016, 18:07:16 UTC |
1a4fdff | Stephan Ewen | 08 November 2016, 16:13:19 UTC | [FLINK-4985] [checkpointing] Report canceled / declined checkpoints to the Checkpoint Coordinator | 08 November 2016, 18:07:16 UTC |