Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/5720#discussion_r176482195
--- Diff:
flink-metrics/flink-metrics-jmx/src/test/java/org/apache/flink/runtime/jobmanager/JMXJobManagerMetricTest.java
---
@@ -92,28 +101,26 @@ public void testJobManagerJMXMetricAccess() throws
Exception {
true),
null));
- flink.waitForActorsToBeAlive();
-
- flink.submitJobDetached(jobGraph);
+ ClusterClient<?> client =
MINI_CLUSTER_RESOURCE.getClusterClient();
+ client.setDetached(true);
+ client.submitJob(jobGraph,
JMXJobManagerMetricTest.class.getClassLoader());
- Future<Object> jobRunning =
flink.getLeaderGateway(deadline.timeLeft())
- .ask(new
TestingJobManagerMessages.WaitForAllVerticesToBeRunning(jobGraph.getJobID()),
deadline.timeLeft());
- Await.ready(jobRunning, deadline.timeLeft());
+ FutureUtils.retrySuccesfulWithDelay(
+ () -> client.getJobStatus(jobGraph.getJobID()),
+ Time.milliseconds(10),
+ deadline,
+ status -> status == JobStatus.RUNNING,
--- End diff --
The check whether the job is running does not necessarily mean that all
vertices are running. But I guess what we are waiting for is the initialization
of the `CheckpointStatsTracker`.
---
