Github user hmcl commented on a diff in the pull request:
https://github.com/apache/storm/pull/1924#discussion_r99931862
--- Diff:
external/storm-kafka-client/src/main/java/org/apache/storm/kafka/spout/KafkaSpout.java
---
@@ -289,6 +293,13 @@ private void doSeekRetriableTopicPartitions() {
kafkaConsumer.seek(rtp,
acked.get(rtp).getCommittedOffset() + 1); // Seek to last committed offset
}
}
+
+ if(!retriableTopicPartitions.isEmpty()) {
+ //Pause other partitions for this poll so
maxUncommittedPartitions isn't exceeded by too much
--- End diff --
I am still not quite understanding why we need to pause all other topic
partitions in the case of a retrial. This could potentially lead to the
scenario that one topic partition may be causing a lot of failures, but all
others are running fine. If that's the case, we will the throttling all other
topic-partitions for no particular reason.
Are you doing this pause to handle the case when `nunUncommittedOffsets`
may exceed `maxUncommittedOffsets`? If that is the case, then in the case of a
retrial we should only only when `nunUncommittedOffsets >=
maxUncommittedOffsets`. Otherwise perhaps we should poll from all
topic-partitions.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
