gyfora commented on code in PR #983:
URL:
https://github.com/apache/flink-kubernetes-operator/pull/983#discussion_r2120321410
##########
flink-kubernetes-operator/src/main/java/org/apache/flink/kubernetes/operator/observer/JobStatusObserver.java:
##########
@@ -132,8 +134,50 @@ protected void
observeJobManagerExceptions(FlinkResourceContext<R> ctx) {
}
var exceptionHistory = history.getExceptionHistory();
- List<JobExceptionsInfoWithHistory.RootExceptionInfo> exceptions =
- exceptionHistory.getEntries();
+ var exceptions = exceptionHistory.getEntries();
+
+ String currentJobId = jobStatus.getJobId();
+ var cacheEntry = ctx.getExceptionCacheEntry();
+
+ if (!cacheEntry.isInitialized()) {
+
+ Instant lastExceptionTs;
+ if (exceptions == null || exceptions.isEmpty()) {
+ // If the job doesn't have any exceptions set to MIN as we
always have to record
+ // the next
+ lastExceptionTs = Instant.MIN;
+ } else {
+ var k8sExpirationTs =
Instant.now().minus(MAX_K8S_EVENT_AGE);
+ var maxJobExceptionTs =
+ exceptions.stream()
+ .map(e ->
Instant.ofEpochMilli(e.getTimestamp()))
+ .max(Comparator.naturalOrder())
+ .orElseThrow();
+
+ if (maxJobExceptionTs.isBefore(k8sExpirationTs)) {
+ // If the last job exception was a long time ago, then
there is no point in
+ // checking in k8s.
+ lastExceptionTs = maxJobExceptionTs;
Review Comment:
there is no config for this (nothing to tune) and the optimization can be
very important when the operator starts up because then the cache is empty and
it would fetch events for every single job. In most cases this filter
completely eliminates that so this greatly reduces the startup api server load
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
