- 05 7月, 2016 7 次提交
-
-
由 Stephan Ewen 提交于
Test utilities include the StreamingMultipleProgramsTestBase and StreamingTestEnvironment. This moves the ITCases for streaming into 'flink-tests' to achieve that. This closes #2092
-
由 Stephan Ewen 提交于
Makes the JUnit test utils (TestLogger, retry rules, ...) properly available to other projects without the 'flink-core' test-jar, via the 'flink-test-utils-junit' project. Makes the ForkableMiniCluster, TestEnvironment, and other test utilities available in the 'main' scope of the 'flink-test-utils' project. Creates a 'flink-test-utils-parent' project that holds the 'flink-test-utils-junit' and 'flink-test-utils' project. Also moves some tests between projects and inlines some very simple utility functions in order to simplify some test jar dependencies.
-
由 Aljoscha Krettek 提交于
Before, job would fail with an exception if the checkpoint filesystem was not accessible from the client. This is preventing some correct jobs from running. Now, only print a warning. This closes #2157
-
由 Josh 提交于
-
由 Josh 提交于
-
由 zentol 提交于
Use distinct non-default port ranges for all tests.
-
由 twalthr 提交于
-
- 04 7月, 2016 7 次提交
-
-
由 Fabian Hueske 提交于
This closes #2126.
-
由 Stefan Richter 提交于
Before, when processing an element that would end up in a late window (when using a MergingWindowAssigner), the element would be added to the MergingWindowSet. After determining that the window is late it would not be removed from the MergingWindowSet. This can lead to problems with other elements being merged into these "phantom" windows and causing triggers to be added for empty windows. This also fixes the same code in EvictingWindowOperator.
-
由 Jark Wu 提交于
This closes #2173.
-
由 Aljoscha Krettek 提交于
Before, this was trying to cast the StreamTaskState directly to a StateHandle<DataInputView> while it should take the operator state, i.e: StateHandle<DataInputView> stateHandle = (StateHandle<DataInputView>) state.getOperatorState();
-
由 Robert Metzger 提交于
-
由 Robert Metzger 提交于
This closes #2108 This closes #2058 because its an invalid pull request.
-
由 Ismaël Mejía 提交于
Some of the changes include: - Remove unneeded dependencies (nano, wget) - Remove apt lists to reduce image size - Reduce number of layers on the docker image (best docker practice) - Remove useless variables and base the code in generic ones e.g. FLINK_HOME - Change the default JDK from oracle to openjdk-8-jre-headless, based on two reasons: 1. You cannot legally repackage the oracle jdk in docker images 2. The open-jdk headless is more appropriate for a server image (no GUI stuff) - Return port assignation to the standard FLINK one: Variable: docker-flink -> flink taskmanager.rpc.port: 6121 -> 6122 taskmanager.data.port: 6122 -> 6121 jobmanager.web.port: 8080 -> 8081 This closes #2176
-
- 02 7月, 2016 7 次提交
-
-
由 Greg Hogan 提交于
-
由 Greg Hogan 提交于
-
由 Greg Hogan 提交于
Adds a GraphAnalytic to replace the checksumHashCode Java and Scala utility functions. This closes #2188
-
由 Greg Hogan 提交于
A delegating GraphAlgorithm wraps a GraphAlgorithm result with a delegating proxy object. The delegated object can be replaced when the same algorithm is run on the same input with a mergeable configuration. This allows algorithms to be composed of implicitly reusable algorithms without publicly sharing intermediate DataSets. This closes #2032
-
由 Maximilian Michels 提交于
This closes #2191
-
由 Maximilian Michels 提交于
This removes the leaderUpdated method from the framework. Further it lets the RM client thread communicate directly with the ResourceManager actor. This is fine since the two are always spawned together. Failures of the ResourceManager actor will lead to dropped messages of the RM client thread. Failures of the RM client thread will inform the JobManager. The leaderUpdated() method was used to signal the ResourceManager framework that a new leader was elected. However, the method was not always called when the leader changed, only when a new leader was elected. This dropped all messages from the async Yarn RM client thread (YarnResourceManagerCallbackHandler) for the time that the old leader had failed and no new leader had been elected. The Yarn RM client thread used leader tagged messages to communicate with the main Flink ResourceManager actor. This closes #2190
-
由 Maximilian Michels 提交于
This fixes a regression of 0483ba58 where the classpath would start off with the application master's home dir.
-
- 01 7月, 2016 18 次提交
-
-
由 zentol 提交于
-
由 Maximilian Michels 提交于
- always ship the lib folder - properly setup the classpath from the supplied ship files - cleanup deploy() method of YarnClusterDescriptor - add test case This closes #2187
-
由 Maximilian Michels 提交于
This closes #2168
-
由 Maximilian Michels 提交于
- lazily initialize ActorSystem - make sure it is not created before job execution - print connection information on the CLI This closes #2189
-
由 Maximilian Michels 提交于
-
由 Maximilian Michels 提交于
-
由 Maximilian Michels 提交于
-
由 Maximilian Michels 提交于
- user specifies no parallelism -> parallelism is adjusted to #taskSlots * #nodes. - user specifies parallelism but no #taskSlots or too few slots -> #taskSlots are set such that they meet the parallelism
-
由 zentol 提交于
-
由 zentol 提交于
-formatting -don't use implicit conversions -fixed ObjectName for mapped bufferpool
-
由 zentol 提交于
-
由 zentol 提交于
-
由 zentol 提交于
-
由 zentol 提交于
This closes #2146
-
由 Aljoscha Krettek 提交于
-
由 Mao Wei 提交于
This closes #2179
-
由 Greg Hogan 提交于
This closes #2185
-
由 Greg Hogan 提交于
Rename MessageFunction to ScatterFunction and VertexUpdateFunction to GatherFunction. Change the parameter order in Graph.runScatterGatherIteration(VertexUpdateFunction, MessagingFunction) to Graph.runScatterGatherIteration(ScatterFunction, GatherFunction) This closes #2184
-
- 30 6月, 2016 1 次提交
-
-
由 Aljoscha Krettek 提交于
Before, when snapshotting, we printed a log message about the file input format not being checkpointable when the current split was "null". Now, we only print the message when when appropriate. This closes #2174
-