- 27 9月, 2017 15 次提交
-
-
由 Aljoscha Krettek 提交于
commons-io is only usable as a transitive dependency of the Hadoop dependencies. We can just use the Java ByteArrayOutputStream and get rid of that dependency.
-
由 Aljoscha Krettek 提交于
-
由 Aljoscha Krettek 提交于
-
由 Aljoscha Krettek 提交于
-
由 Aljoscha Krettek 提交于
There are methods for this in flink-hadoop-compatibility.
-
由 Aljoscha Krettek 提交于
This removes all Hadoop-related methods from ExecutionEnvironment (there are already equivalent methods in flink-hadoop-compatibility (see HadoopUtils and HadoopInputs, etc.). This also removes Hadoop-specific tests from flink-tests because these are duplicated by tests in flink-hadoop-compatibility. This also removes Hadoop-specic example code from flink-examples: the DistCp example and related code.
-
由 zhangminglei 提交于
This closes #4659.
-
由 Tony Wei 提交于
This closes #4679.
-
由 yew1eb 提交于
This closes #4694.
-
由 Tzu-Li (Gordon) Tai 提交于
Introduces a common test base that for all REST responses, a subclass should be implemented to verify that the response can be correctly marshalled and unmarshalled. This closes #4691. This closes #4720.
-
由 Tzu-Li (Gordon) Tai 提交于
The original naming wouldn't make sense for the FLIP-6 redesign, since we would have multiple per-job JobManagers for each cluster, which shares the same configuration. The REST path is still left untouched and not part of this commit, as that would involve more changes in flink-runtime-web.
-
由 Tzu-Li (Gordon) Tai 提交于
This commit lets the JobManagerConfigHandler implement the LegacyRestHandler interface in order to be ported to the new REST endpoint. This includes the introduction of ClusterConfiguration response body and ClusterConfigurationHeaders. The DispatcherRestEndpoint now also registers the JobManagerConfigHandler.
-
由 Bang, Phiradet 提交于
This closes #4704.
-
由 Bowen Li 提交于
[FLINK-7393] [kinesis] Move unit tests of KinesisConfigUtil from FlinkKinesisConsumerTest to KinesisConfigUtilTest This closes #4708.
-
由 David Anderson 提交于
This closes #4687.
-
- 26 9月, 2017 8 次提交
-
-
由 twalthr 提交于
-
由 Xpray 提交于
This closes #4658.
-
由 twalthr 提交于
-
由 Till Rohrmann 提交于
FlinkFutureException was introduced to fail a CompletableFuture callback. However, there was already such a class which allows to better handle failures in different stages which is the java.util.CompletionException. Therefore we replace FlinkFutureException by CompletionException and remove the former. This closes #4701.
-
由 twalthr 提交于
-
由 Jark Wu 提交于
This closes #4536.
-
由 Till Rohrmann 提交于
This reverts commit f1b2b83d because it makes the OperatorSubtaskState unserializable due to the usage of guava.Multimap.get. This effectively breaks the master.
-
由 Till Rohrmann 提交于
Ports the CurrentJobsOverviewHandler to the new REST endpoint by letting it implement the LegacyRestHandler interface. This commit changes the JobDetails JSON such that it now contains the number of tasks for each ExecutionState, including SCHEDULED, DEPLOYING, CREATED and RECONCILING. These state will now also be displayed in the web frontend. Change MultipleJobsDetails to store a Collection<JobDetails> instead of JobDetails[] Use MultipleJobsDetails#FIELD_NAME_ for serialization in CurrentJobsOverviewHandler This closes #4688.
-
- 25 9月, 2017 8 次提交
-
-
由 Piotr Nowojski 提交于
This is not complete refactor, some methods still relay on the order of the new and old operators.
-
由 Piotr Nowojski 提交于
-
由 Stefan Richter 提交于
[FLINK-7619] Improved abstraction of AbstractAsyncIOCallable to better fit the current usage pattern.
-
由 Stefan Richter 提交于
-
由 Aljoscha Krettek 提交于
-
由 Bowen Li 提交于
-
由 Aljoscha Krettek 提交于
Recently, the aligned window operators were removes, these classes where leftover after that removal.
-
由 Bowen Li 提交于
-
- 24 9月, 2017 1 次提交
-
-
由 Aljoscha Krettek 提交于
-
- 23 9月, 2017 1 次提交
-
-
由 Fabian Hueske 提交于
[FLINK-7656] [runtime] Switch to user classloader before calling initializeOnMaster and finalizeOnMaster. This closes #4690.
-
- 22 9月, 2017 5 次提交
-
-
由 desktop 提交于
This closes #4685.
-
由 Bowen Li 提交于
The updateCredentials delay is an ignorable warning that occurs due to the fact that the default credential update delay is longer than the await termination timeout when shutting down KPL. See https://github.com/awslabs/amazon-kinesis-producer/issues/10 for details. This closes #4657.
-
由 Bowen Li 提交于
KinesisProducerLibrary (KPL) 0.10.x had been using a One-New-Thread-Per-Request model for all requests sent to AWS Kinesis, which is very expensive. 0.12.4 introduced a new ThreadingMode - Pooled, which will use a thread pool. This hugely improves KPL's performance and reduces consumed resources. By default, KPL still uses per-request mode. We should explicitly switch FlinkKinesisProducer's KPL threading mode to 'Pooled'. This closes #4656.
-
由 Mikhail Lipkovich 提交于
-
由 Till Rohrmann 提交于
This commit changes the fencing behaviour such that a component which is not the leader will set its fencing token to null. This distinction allows to throw different exceptions depending on whether it is a token mismatch or whether the receiver has no fencing token set (== not being the leader). This closes #4689.
-
- 21 9月, 2017 2 次提交
-
-
由 Aljoscha Krettek 提交于
This also adds an end-to-end test that verifies correct order for both classes and resources.
-
由 Fabian Hueske 提交于
This closes #4635.
-