Allow Flink to ignore savepoint on restore if the states of the old and new jobs are incompatible #589
Description
We have a streaming job that runs with JobSpec.takeSavepointOnUpdate: true property. When updating the job and restoring it from savepoint, we need to have a way to gracefully handle the failure of the restore process if the new job state and the previous job state are incompatible.
We would like the operator to:
- Make an attempt to restore from savepoint
- If it does not succeed, ignore savepoint and start the job with the default state.
Since there are cases for which this approach is not acceptable, it would be good to have an option to configure the desired behaviour.