Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/4734#discussion_r143716493
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/dispatcher/Dispatcher.java
---
@@ -242,6 +248,86 @@ public void start() throws Exception {
}
@Override
+ public CompletableFuture<JobStatusesWithIdsOverview>
requestJobIdsOverview(@RpcTimeout Time timeout) {
+ final int numJobs = jobManagerRunners.size();
+
+ ArrayList<CompletableFuture<Tuple2<JobID, JobStatus>>>
jobStatuses = new ArrayList<>(numJobs);
+ for (Map.Entry<JobID, JobManagerRunner> jobManagerRunnerEntry :
jobManagerRunners.entrySet()) {
+ CompletableFuture<JobStatus> jobStatusFuture =
+
jobManagerRunnerEntry.getValue().getJobManagerGateway().requestJobStatus(timeout);
+
+ jobStatuses.add(jobStatusFuture.thenApply(jobStatus ->
Tuple2.of(jobManagerRunnerEntry.getKey(), jobStatus)));
+ }
+
+ CompletableFuture<Collection<Tuple2<JobID, JobStatus>>>
combinedJobStatusesFuture = FutureUtils.combineAll(jobStatuses);
+
+ return combinedJobStatusesFuture.thenApply(
+ jobStatusesWithIds -> {
+ List<JobID> jobsCreated = new LinkedList<>();
+ List<JobID> jobsRunning = new LinkedList<>();
+ List<JobID> jobsFinished = new LinkedList<>();
+ List<JobID> jobsCancelling = new LinkedList<>();
+ List<JobID> jobsCancelled = new LinkedList<>();
+ List<JobID> jobsFailing = new LinkedList<>();
+ List<JobID> jobsFailed = new LinkedList<>();
+ List<JobID> jobsRestarting = new LinkedList<>();
+ List<JobID> jobsSuspended = new LinkedList<>();
+ List<JobID> jobsReconciling = new
LinkedList<>();
--- End diff --
Not entirely sure, but I think that `ArrayList` would be better here. Even
though we will most likely only add few elements per list, a linked list is
probably a bit slower since `LinkedList` uses some more complex internal data
structures.
---
