tillrohrmann commented on a change in pull request #6751: [FLINK-10403] Port
JobManagerHAProcessFailureBatchRecoveryITCase to new code base
URL: https://github.com/apache/flink/pull/6751#discussion_r223208480
##########
File path:
flink-tests/src/test/java/org/apache/flink/test/recovery/JobManagerHAProcessFailureBatchRecoveryITCase.java
##########
@@ -141,15 +151,14 @@ public
JobManagerHAProcessFailureBatchRecoveryITCase(ExecutionMode executionMode
*/
private void testJobManagerFailure(String zkQuorum, final File
coordinateDir, final File zookeeperStoragePath) throws Exception {
Configuration config = new Configuration();
- config.setString(CoreOptions.MODE, CoreOptions.LEGACY_MODE);
config.setString(HighAvailabilityOptions.HA_MODE, "ZOOKEEPER");
config.setString(HighAvailabilityOptions.HA_ZOOKEEPER_QUORUM,
zkQuorum);
config.setString(HighAvailabilityOptions.HA_STORAGE_PATH,
zookeeperStoragePath.getAbsolutePath());
ExecutionEnvironment env =
ExecutionEnvironment.createRemoteEnvironment(
"leader", 1, config);
env.setParallelism(PARALLELISM);
- env.setNumberOfExecutionRetries(1);
+ env.setRestartStrategy(RestartStrategies.fixedDelayRestart(1,
0L));
env.getConfig().setExecutionMode(executionMode);
Review comment:
The `ExecutionMode` defines the data exchange between producers and
consumers. When using `BATCH`, then the producers (e.g. input for a reduce
operation) first need to finish before the consumer is started. In `PIPELINED`
mode, the consumer will start as soon as a single producer has produced some
consumable data.
The test job contains a shuffle operation (between mapper and reducer) which
is affected by this setting.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
