Trident Topology Hang

[Adam Lewis <ma...@adamlewis.com>]

I am having an issue with a trident topology that is "hanging"
(deadlocking?) when I add a multiReduce "branch".  I've run into this issue
upon assembling this from two smaller topologies which both work fine.  I'm
working on reproducing this in some simplified code, but doing so isn't
trivial.  In the meantime, I'm wondering if there are any issues that are
apparent from looking at the attached diagram of my topology.

Specifically, the red "aggregate" step gets tuples added to it, but never
has its "complete" method called.  If I remove the yellow "MultiReduce"
step (and steps downstream from it), the problem goes away and remainder of
the topology runs fine top to bottom.

Are there any subtleties about how trident synchronizes the multi reducers
which could cause this deadlock?  I don't have any cycles and the topology
submits fine, but for some reason never completes the aggregation.

Thanks,
Adam