- 28 5月, 2019 1 次提交
-
-
由 Huang Xingbo 提交于
This closes #8472
-
- 27 5月, 2019 6 次提交
-
-
由 yanghua 提交于
-
由 Timo Walther 提交于
This closes #8510.
-
由 Timo Walther 提交于
-
由 Chesnay Schepler 提交于
Replicate jdk.tools exclusion in every module that requires them. Remove exclusion from root pom to prevent side-effects.
-
由 leesf 提交于
-
由 Bo WANG 提交于
-
- 26 5月, 2019 2 次提交
- 25 5月, 2019 3 次提交
-
-
由 Abdul Qadeer (abqadeer) 提交于
This closes #8154.
-
由 godfreyhe 提交于
[FLINK-12575][table-planner-blink] Introduce planner rules to remove redundant shuffle and collation This closes #8499 i#refactor satisfyTraits method
-
由 Yun Tang 提交于
This closes #8391.
-
- 24 5月, 2019 22 次提交
-
-
由 Gen Luo 提交于
This closes #8402
-
由 Dian Fu 提交于
This closes #8420
-
由 Dawid Wysakowicz 提交于
In this commit both registration & tables lookup go through CatalogManager with respect to the current default catalog & database. It is not decoupled from Calcite yet. The next step will be to register the CatalogTables built exclusively in the table api rather than converted already to the Calcite's Tables. This closes #8404
-
由 bowen.li 提交于
[FLINK-11476] [SQL/TABLE] Create CatalogManager to manage multiple catalogs and encapsulate Calcite schema
-
由 sunjincheng121 提交于
This is closes #8528
-
由 Biao Liu 提交于
-
由 Chesnay Schepler 提交于
-
由 Chesnay Schepler 提交于
-
由 azagrebin 提交于
This part of Shuffle API refactoring: make task not depend on the concrete implementation of ResultPartitionWriter (ResultPartition).
-
由 时无两丶 提交于
-
由 sunjincheng121 提交于
This closes #8525
-
由 Rui Li 提交于
This PR adds support for Hive partitions in HiveCatalog. This closes #8449.
-
由 bowen.li 提交于
This PR creates HiveCatalogFunction class and adds support for Hive function operations in HiveCatalog. This closes #8507.
-
由 Zhijiang 提交于
If the consumer can not establish a connection to remote task executor while requesting remote subpartition, which might indicate the remote task executor is not reachable. We could wrap this connection exception into new proposed PartitionConnectionException which also extends PartitionException, then the job master would decide whether to restart the upstream region to re-producer partition data. This closes #8509.
-
由 Zhijiang 提交于
The new proposed PartitionException would cover all the cases of consuming partition failure which causes consumer failed, then JM decides to restart the producer based on this exception. [FLINK-6227][network] (part 2)Make current PartitionNotFoundException extend PartitionException This closes #8242.
-
由 Zhijiang 提交于
The current usage of PartitionException is only for describing the RPC of update parttitions failed, so the exception rename would not have any other effects. In flip1 PartitionException is used for presenting all the cases that indicate to restart the producer while the consumer failed.
-
由 Zhijiang 提交于
It is necessary for flip1 to make sure the PartitionNotFoundException would be setted on the RemoteInputChannel while retriggering partition request, so a new unit test is added to cover this case.
-
由 Zhijiang 提交于
[hotfix][network] Introduce PartititonRequestClient interface for creating simple client instance in tests
-
由 Zhijiang 提交于
It is necessary for flip1 to make sure the PartitionNotFoundException would be thrown by LocalInputChannel#requestSubpartition if the partition was not registered in ResultPartitionManager before. So a new unit test is added to cover this case.
-
由 Zhijiang 提交于
It is necessary to make sure the network server would not transform or swallow the PartitionNotFoundException thrown by ResultPartitionManager, so a new unit test is added for PartitionReuqestServerHandler to cover this case.
-
由 Zhijiang 提交于
It is necessary to make sure PartitionNotFoundException is thrown on producer side for flip1, so some new tests are added to cover the cases of creating subpartition view via ResultPartitionManager for registered/unregistered partitions.
-
由 zhijiang 提交于
-
- 23 5月, 2019 6 次提交
-
-
由 Gary Yao 提交于
Add new SchedulerNG stub implementation, which will represents the future default scheduler. Add feature toggle to switch between existing scheduler and stub implementation. Add ThrowingRestartStrategy to validate that in new scheduling code paths, the legacy restart strategies are not used. This closes #8452.
-
由 Gary Yao 提交于
Move SchedulerNG, SchedulerNGFactory, and its implementations to package org.apache.flink.runtime.scheduler.
-
由 Gary Yao 提交于
-
由 Gary Yao 提交于
-
由 Timo Walther 提交于
-
由 Timo Walther 提交于
-