I'm trying to run some very simple examples with kafka and running into a strange problem.
I've documented this in detail in https://issues.apache.org/jira/browse/KAFKA-12594#. The linked repo - https://github.com/jamii/streaming-consistency/tree/main/kafka-streams - should be runnable on any linux and only depends on nix. The short version is: * I'm reading a table from a single topic with json values * If I write that table directly back to kafka then I can see the results in the output topic, and a `.mapValues(...println...)` on that stream will show results in the terminal * (On some runs, a continguous subset of ~500 input records don't appear in the output) * If I uncomment any of the other examples, they don't produce any output, the write I mentioned in the above bullet also doesn't work and the `.mapValues(...println...)` no longer produces any results * There are no errors in the kafka logs (the full logs are attached in one of the comments on the issue) * The process doesn't terminate or crash. * Using strace I can see that the demo process reads the entire input topic, even though none of the printlns fire. The symptoms above are reliably reproduced on my laptop (nixos) and on a fresh ubuntu 20.10 vm on vultr, with one exception - I once saw the sums topic produce records on vultr. That it worked at least once strongly suggests to me that there is a race somewhere, rather than eg a silent logic bug in the examples. But I don't have enough knowledge of the kafka streams internals to track it down any further.
