Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/997#issuecomment-129556271
I agree with @chenliang613 , it would be great to have a brief description
how this mechanism is implemented.
I think there is also a problem still, because InputSplits are assigned
once again after the dataflow graph (ExecutionGraph) is restarted. Since their
assignment is lazy (to better load balance), they may get distributed in a
different way on the re-try, compared to the original execution.
To fix this, the input split assignment would need to be part of the
checkpoints as well.
---
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.
---
