- 28 2月, 2020 22 次提交
-
-
由 Yangze Guo 提交于
-
由 Zhu Zhu 提交于
[FLINK-16180][runtime] Replace the nullable vertexExecution in ScheduledUnit with a non-null executionVertexId
-
由 Zhu Zhu 提交于
-
由 Zhu Zhu 提交于
-
由 dianfu 提交于
[FLINK-16271][python] Introduce ArrowPythonScalarFunctionOperator for vectorized Python UDF execution
-
由 Robert Metzger 提交于
This closes #11241
-
由 Robert Metzger 提交于
-
由 Robert Metzger 提交于
-
由 Piotr Nowojski 提交于
This doesn't fully support input selection, watermarks, latency markers and keyed contextes yet
-
由 Piotr Nowojski 提交于
This creates new hierarchy of task harnesses that could be used for other tasks as well.
-
由 Piotr Nowojski 提交于
-
由 Piotr Nowojski 提交于
-
由 Piotr Nowojski 提交于
This is going to be used for runtime support of MultipleInputStreamOperator
-
由 Piotr Nowojski 提交于
-
由 Piotr Nowojski 提交于
-
由 Piotr Nowojski 提交于
[FLINK-16060][checkpointing] Generalize LinkedBufferStorage to support arbitrary number of linked buffer storages
-
由 Piotr Nowojski 提交于
-
由 Piotr Nowojski 提交于
-
由 Piotr Nowojski 提交于
-
由 Piotr Nowojski 提交于
-
由 Piotr Nowojski 提交于
This interface is still not complete - it's missing for example watermark and latency markers support.
-
由 Jark Wu 提交于
-
- 27 2月, 2020 16 次提交
-
-
由 wangyang0918 提交于
The warning log should show up when the JobManager/TaskManager memory is not a multiple of YARN minimum allocation memory. Because YARN will always normalize the resource request by insuring that the requested memory is a multiple of minimum allocation. The minimum allocation could be configured in YARN ResourceManager via 'yarn.scheduler.minimum-allocation-mb'. This closes #11176.
-
由 Zhijiang 提交于
The method SingleInputGate#getConsumedResultId is never used, so remove it to cleanup the codes.
-
由 Zhijiang 提交于
The ResultPartitionID in AddCredit message is never used on upstream side, so we can remove it to cleanup the codes. There would have another two benefits to do so: 1. Reduce the total message size from previous 52 bytes to 20 bytes. 2. Decouple the dependency with InputChannel#getPartitionId.
-
由 Hequn Cheng 提交于
-
由 ptmagic 提交于
The option was deprecated in 1.10 in favour of `taskmanager.memory.process.size`. This closes #10890.
-
由 Yangze Guo 提交于
-
由 Yangze Guo 提交于
-
由 huangxingbo 提交于
[FLINK-15972][python][table-planner][table-planner-blink] Add Python building blocks to make sure the basic functionality of Python TableFunction could work This close #11130.
-
由 huangxingbo 提交于
-
由 huangxingbo 提交于
-
由 Rui Li 提交于
-
由 Rui Li 提交于
-
由 cpugputpu 提交于
This closes #11187.
-
由 kevin.cyj 提交于
Before this change, in network benchmark (for example 1000 channels benchmark with 4 record writers) StreamNetworkBenchmarkEnvironment#createInputGate was creating 1000 input gates with 4 input channels each, which doesn't make much sense. This commit is changing that to a single receiver with 4 input gates and each with 1000 channels. It is achieved by providing testing implementations of InputChannels, which are using channel index for requesting subpartitions from ResultPartition, instead of subpartition index. Thanks to that, we can map a single ResultPartition with N subpartitions, to a single instance of InputGate with N channels. The change also influences the benchmark results, overall, the performance goes down a bit because of the decrease of floating buffers and the followings are benchmark results before and after this change: ------------------------------------------------------------------Before---------------------------------------------------------------------- Benchmark (channelsFlushTimeout) Mode Cnt Score Error Units DataSkewStreamNetworkThroughputBenchmarkExecutor.networkSkewedThroughput N/A thrpt 30 17079.534 ± 830.532 ops/ms StreamNetworkBroadcastThroughputBenchmarkExecutor.networkBroadcastThroughput N/A thrpt 30 599.664 ± 13.325 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 100,100ms thrpt 30 45629.898 ± 1623.455 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 100,100ms,SSL thrpt 30 9817.421 ± 216.075 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 1000,1ms thrpt 30 25442.152 ± 968.340 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 1000,100ms thrpt 30 27944.285 ± 518.106 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 1000,100ms,SSL thrpt 30 7820.549 ± 895.862 ops/ms StreamNetworkLatencyBenchmarkExecutor.networkLatency1to1 N/A avgt 30 13.184 ± 0.093 ms/op ------------------------------------------------------------------After----------------------------------------------------------------------- Benchmark (channelsFlushTimeout) Mode Cnt Score Error Units DataSkewStreamNetworkThroughputBenchmarkExecutor.networkSkewedThroughput N/A thrpt 30 17345.574 ± 370.647 ops/ms StreamNetworkBroadcastThroughputBenchmarkExecutor.networkBroadcastThroughput N/A thrpt 30 608.881 ± 12.054 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 100,100ms thrpt 30 41732.518 ± 1109.436 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 100,100ms,SSL thrpt 30 9689.525 ± 202.895 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 1000,1ms thrpt 30 24106.705 ± 2952.364 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 1000,100ms thrpt 30 27509.665 ± 3246.965 ops/ms StreamNetworkThroughputBenchmarkExecutor.networkThroughput 1000,100ms,SSL thrpt 30 7691.287 ± 927.775 ops/ms StreamNetworkLatencyBenchmarkExecutor.networkLatency1to1 N/A avgt 30 12.758 ± 0.147 ms/op
-
由 HuangXingBo 提交于
-
由 Jingsong Lee 提交于
[FLINK-16265][table][csv] CsvTableSinkFactoryBase should compare LogicalTypes instead of TableSchema (#11229)
-
- 26 2月, 2020 2 次提交
-
-
由 Robert Metzger 提交于
Problem: The logging pattern recently changed, that's why the extaction of port / ip failed This closes #11222
-
由 Robert Metzger 提交于
The problem was that YARN was decommissioning NodeManagers because of low disk space.
-