hntd187 commented on issue #1544: URL: https://github.com/apache/arrow-datafusion/issues/1544#issuecomment-1009645136
@realno So I can update my initial comment to flesh out that design as we walk through it. 1. For delivery guarantees, it seems to make sense to do what Spark did which is exactly once delivery. Idempotent sinks, write ahead logs and checkpointing to ensure the offsets of the stream are always managed. This might be over kill for a first pass and maybe we don't have to initially recover from all failure scenarios but it seems like a good point to get to eventually. 2. For out of order delivery, a first pass to support watermarking seems at the very least necessary. A lazy solution beyond that would be to tell the user to "deal with the possibility" which I think is fine for an MVP level implementation but we want to do better than that. @realno any suggestions I'm not aware of how other streaming engines handle the out of order possibility without doing some research. -- 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]
