For stateful/non-deterministic UDFs, we do not evaluate them in the
optimizer stage. For deterministic UDFs, each invocation should return the
same result.
Before Spark 2.3 release, we assume all the UDFs are deterministic and
stateless. In the recent release Spark 2.3, we allow users to mark the
+1
Tom, thanks for helping this!
Xiao
2018-06-07 9:40 GMT-07:00 Marcelo Vanzin :
> Took a look at our branch and most of the stuff that is not already in
> 2.2 are flaky test fixes, so +1.
>
> On Wed, Jun 6, 2018 at 7:54 AM, Tom Graves
> wrote:
> > Hello all,
> >
> > I think its time for anoth
