Hi Stephan,
Thank you so much for all these practical suggestions.
I agree that hybrid batch/streaming execution is an awesome idea and that
no checkpoint could be successful during the batch phase. However, the
checkpoint could fail frequently because of the large batch size, which
might be
Before jumping into the designs of other mechanisms, can we clarify whether
adjusting checkpointing is really the right approach here?
What about storing empty state in the checkpoints, so that any recovery
simply does a full replay of the input.
Regarding the hybrid batch/streaming execution:
Hi Stephan,
Thank you so much for replying and suggesting! Follow by your question, I
would give some explanation and new thoughts.
(1) More detailed info about CDC use case.
In the previous design of FLINK-CDC, they would start a full-table scanning
at the beginning by holding a read-write
Hi!
Thanks for writing this FLIP, and interesting idea.
I would like to understand a bit better why exactly we need this, and what
our alternative options are. My main concerns are the following:
*(1) Can we achieve this without changing the checkpointing mechanism?*
The checkpoint mechanism
Hi Piotrek,
Thanks for your feedback!
1. Why not ExternallyInducedSourceReader/ExternallyInducedSource?
a. The
`org.apache.flink.api.connector.source.ExternallyInducedSourceReader` and
`org.apache.flink.api.connector.source.ExternallyInducedSource` seems like
playing the role of checkpoint
Hi Senhong,
Thanks for the proposal. I have a couple of questions.
Have you seen
`org.apache.flink.streaming.api.checkpoint.ExternallyInducedSource` (for
the legacy SourceFunction) and
`org.apache.flink.api.connector.source.ExternallyInducedSourceReader` (for
FLIP-27) interfaces? They work the
Here is some brief context about the new feature.
1. Actively checkpoint rejecting by the operator. Follow by the current
checkpoint mechanism, one more preliminary step is added to help the
operator determine that if it is able to take snapshots. The preliminary
step is a new API provided to the
Hi guys,
We would like to start a discussion on the new FLIP about rejecting
checkpoints on the operator level. The basic idea is to allow the operator
to reject a checkpoint when it is not under a proper situation and returning
a proper failure reason.