Hello Avi, Yes, this is possible: although we checked nullable keys when doing reduce / aggregations:
https://github.com/apache/kafka/blob/0.10.0/streams/src/main/java/org/apache/kafka/streams/kstream/internals/KStreamReduce.java#L67 We do not check if the there are any values returned from the underlying state store, and when we cannot find such a match, null is returned to deserializer. I think in general the library should guard this case instead of letting desers worry about it. Do you want to file a JIRA reporting this bug so we can follow-up? Guozhang On Fri, Jun 10, 2016 at 11:27 AM, Avi Flax <avi.f...@parkassist.com> wrote: > > > On Jun 10, 2016, at 14:24, Avi Flax <avi.f...@parkassist.com> wrote: > > > > Hi, I’m using Kafka Streams (0.10.0) with JRuby, most of my > scripts/nodes are working well at this point, except for one which is using > reduceByKey. > > Whoops, I should probably share my code as well! > > Here’s the topology: > > > builder.stream(key_serde, val_serde, 'visit-update-events') > .reduceByKey(-> (a, b) { a.merge b }, key_serde, val_serde, > 'intermediate-visits') > .to(key_serde, val_serde, 'visits’) > > > This is using Ruby syntax, but hopefully it’s fairly readable. I’ve added > it to this gist as well: > https://gist.github.com/aviflax/3428cdbaa18aca9bf0a958c6d5eac2bd > > Thanks! > Avi -- -- Guozhang
