Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/6283#discussion_r202103545
--- Diff:
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/CoordinatorShutdownTest.java
---
@@ -83,6 +86,10 @@ public void testCoordinatorShutsDownOnFailure() {
CheckpointRetentionPolicy.NEVER_RETAIN_AFTER_TERMINATION,
true),
null));
+
+ ExecutionConfig executionConfig = new ExecutionConfig();
+
executionConfig.setRestartStrategy(RestartStrategies.fallBackRestart());
--- End diff --
You're right. I'm just wondering whether you ever want to enable
checkpointing without a restart strategy. So to speak if you set
`FallbackRestartStrategy`, enable checkpointing and set `NoRestartStrategy` as
the server side `RestartStrategy`, then do you want `FixedRestartStrategy` or
`NoRestartStrategy`?
On the other hand you might want to disable restarting for all jobs running
on your cluster by setting the restart strategy to `NoRestartStrategy`.
Maybe the proper solution would be to set `ExecutionConfig#restartStrategy`
to `FallbackRestartStrategy` and introduce a new default server side restart
strategy `NoOrFixedIfCheckpointingEnabled` which resolved to
`FixedRestartStrategy` if checkpointing is enabled and if not it resolves to
`NoRestartStrategy`.
What do you think?
---
