VTK Executive model is bogus in parallel for structured data
This issue was created automatically from an original Mantis Issue. Further discussion may take place here.
Here is the description of the expected behavior of the executive.
In case of execution, parallel or not, if a filter fails (RequestData return 0), the executive should NOT try to re-execute it unless the MTime of the filter has change or if its input has changed.
Currently, a second execution happen because the executive thinks that a different extent request has been made, while in fact, the only reason for that was that no data was produced during the first pass and therefore the data extent and the request extent still don't match.
---- Moreover ----
In case of extent smaller than the number of partitions that needs to be made, the executive will decide to not execute on some of the processes and if the filter rely on MPI to exchange some informations, we run into dead locks.
Although, this use case will only show up in case of failure when (RequestData return 0) as the first pass will ALWAYS occurs on all the processes due to the MTime.
The purpose of that bug is not to try to solve it right away but mostly to keep those issue in mind when designing/improving the pipeline implementation.