Github user koeninger commented on the pull request:
https://github.com/apache/spark/pull/4805#issuecomment-78053440
I am now confused about what the purpose of this PR is. The jira seemed to
indicate that the problem was "several third-party offset monitoring tools fail
to monitor the status of Kafka consumer", and that the solution was just to
duplicate offsets into ZK, not to use ZK as the system of record for offsets.
Using ZK as the system of record for offsets means that transactional
exactly-once semantics is impossible. I wouldn't be in favor of that change.
I really think the best way to handle this issue is to make it easier for
users to interact with Kafka-managed offsets. For instance, if the kafka
cluster code proves to be stable, exposing methods to get and set the consumer
offsets. That way people can choose the semantics they need, by saving offsets
to ZK at the point in the job that makes sense for their uses.
That's what we do with our jobs - store offsets into the system of record,
then duplicate them into ZK so that existing monitoring tools are happy.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
