Hi

In our streaming pipeline, we have two pubsub sources.
PCollection<OurType> A and PCollection<OurType> B is created from the two
pubsub sources respectively.

In this case, both A and B is unbounded PCollection. Can we do something
like:
PCollection<OurType> C = PCollectionList.of(A).and(B).apply(Flatten.
pCollections()) ?

When we try to flatten them into one PCollection like above, basically the
pipeline stopped working by repeatedly throwing message "GetData failed:
status: APPLICATION_ERROR(3): Computation S6does not have state family S4
for value read".  Anyone know what this error msg mean?

Alternatively, if we transform A and B into KV<OurType, Integer> and
flatten them into C and do a GroupByKey with a 1 second window then
transform it back to PCollection<OurType>, the pipeline seems working
correctly.

-- 
Richard Tian

Software Engineer
Snap Inc.

Reply via email to