John Spray

This reproduced here: https://buildkite.com/redpanda/redpanda/builds/18567#0184773a-2bce-4e1f-929c-ced32af30a9c

This increasingly looks like it could be a franz-go bug. In that last failure, I can see: - The "Calling PollFetches" log on the client shows all the consumed offsets...

Clearly it's surprising/interesting that this issue is coming up in the tiny smoke test rather than elsewhere (e.g. in scale tests). I wonder if it's because in this case the...

Let's see if this will repro with trace logging https://github.com/redpanda-data/redpanda/pull/7285

#7285 ran the test 300 times without failing. We have nothing to go on, and franz-go has been upgraded since the last this happened, so closing this for now: if...

So, on this instance: https://ci-artifacts.dev.vectorized.cloud/redpanda/0185e491-1357-40f9-b7cb-05b18d5b8221/vbuild/ducktape/results/2023-01-24--001/report.html It is specifically partitions on docker-rp-6 that the client is not seeing data for. Grepping server side for requests + connections over the last few...

This is presumed to be same/similar to https://github.com/redpanda-data/redpanda/issues/6399

Assigning myself as the next action here is likely to be watching+waiting for reproduction with https://github.com/redpanda-data/redpanda/pull/8460

Travis also had a good suggestion to add a background goroutine in kgo-verifier that panics the process (to get backtraces) if the client does not complete within expected timeout, perhaps...

`sev/low` because it's believed to be a client bug.