Runners are responsible for retry semantics, they should catch the failure and choose whether they want to retry or not. I think your reading of the code is correct.
Some I/O layers do retry but that is more about attempting to continue processing within a bundle instead of failing and having the runner choose to retry the bundle or not. On Thu, Aug 30, 2018 at 5:57 AM Etienne Chauchot <[email protected]> wrote: > Hi all, > > I have a question concerning retrial of sources. I've looked at the code > of direct runner and spark runner on bounded sources. As far as I can tell, > if there is a failure in reading a record from the reader of the source, > there will be no retrial from the runner, there will just be an exception > thrown. The only retrial that can take place with the source is the one > that could be done by the reader itself. > > Can you confirm that I did not miss something? > > Etienne >
