Mark and JoeW's comments make sense to me. I think the default should be the safest for the user and should follow the Principle of Least Astonishment[1]. In a flow building UI it's natural to connect things together, so not connecting is the special case.
That said, as Mark pointed out there are some cases where the processor provides a relationship for testing, auditing, etc. and does not require a connection. In those cases the Relationship class' new AutoTerminate property should eliminate the need to set AutoTerminate on a relationship that's usually not going to be connected. After that is propagated out to the processors as appropriate I think this will be less of a problem. [1] https://en.wikipedia.org/wiki/Principle_of_least_astonishment On Mon, Apr 25, 2016 at 8:30 AM, Joe Witt <joe.w...@gmail.com> wrote: > I agree Mark. We should have the user make a simple explicit > expression of intent to terminate data but we should not by default > silently do so. I can see how during early phases of building a brand > new flow or experimenting with NiFi one could feel like they'd rather > it just terminated data by default but I think that preference shifts > over time. > > On Mon, Apr 25, 2016 at 8:23 AM, Mark Payne <marka...@hotmail.com> wrote: > > Manish, > > > > I think changing the default to auto terminate would be a rather > dangerous > > move. When > > a user is creating a dataflow, it would be very easy to overlook one of > the > > relationships > > on a Processor and forget to configure it. If it were configured to > > auto-terminate, the data > > that is routed to that relationship would be deleted, resulting in > > unintentional data loss. > > > > This is why we chose to force the user to explicitly acknowledge that > they > > want a Relationship > > to be auto-terminated. As Pierre noted, there is a ticket [1] that was > > resolved that allows the > > developer to auto-terminate a specific Relationship. However, this > should be > > used sparingly, > > for cases where it is expected that the data routed to that relationship > > will not be needed most > > of the time. > > > > An example of where this may come into play is the 'original' > Relationship > > of the > > MergeContent Processor. There are a few cases where we want the original > > FlowFiles that > > were merged together, so the 'original' Relationship does exist, but in > most > > cases we only > > want the merged FlowFile. So auto-terminating the 'original' > Relationship by > > default may be > > a reasonable thing to do. > > > > Do others have thoughts on the issue? > > > > Thanks > > -Mark > > > > > > [1] https://issues.apache.org/jira/browse/NIFI-1003 > > > > > > > > On Apr 24, 2016, at 6:03 AM, Pierre Villard <pierre.villard...@gmail.com > > > > wrote: > > > > Hi, > > > > I may be wrong, but I think this is something that has been already > > discussed for 1.0 branch. However, I had a quick look on JIRA and didn't > > find anything... > > Otherwise there is NIFI-1003 to allow a default auto-termination for some > > relationships. > > > > Pierre > > > > > > 2016-04-24 11:00 GMT+02:00 Oleg Zhurakousky < > ozhurakou...@hortonworks.com>: > >> > >> Manish > >> > >> That is an interesting suggestion. I personally don't see issues with it > >> and generally do believe it would improve user experience, but > interested > >> what others think. > >> > >> Cheers > >> Oleg > >> > >> Sent from my iPhone > >> > >> On Apr 24, 2016, at 08:46, Manish Gupta 8 <mgupt...@sapient.com> wrote: > >> > >> Hi, > >> > >> > >> > >> Does it make sense to keep all the out-flowing relationships > >> auto-terminated by default when a new processor is dragged in? When user > >> connects the processor and specify a relationship, only the selected one > >> becomes non-terminating. > >> > >> > >> > >> I think this will be good from usability point of view. > >> > >> > >> > >> Thanks, > >> > >> Manish > >> > >> > >> > >> > > > > > > >
