redpanda icon indicating copy to clipboard operation
redpanda copied to clipboard

tests: add assertions & logging to verifiable producer

Open jcsp opened this issue 2 years ago • 1 comments

Cover letter

This is to narrow down a spooky failure seen when the consumer is apparently not waiting for a high enough offset, due to seeing a lower offset in await_consumed_offsets than the VerifiableProducer had earlier reported.

Related: https://github.com/redpanda-data/redpanda/issues/3450

Backport Required

  • [x] not a bug fix
  • [ ] papercut/not impactful enough to backport
  • [ ] v22.2.x
  • [ ] v22.1.x
  • [ ] v21.11.x

UX changes

  • None

Release notes

  • none

jcsp avatar Aug 09 '22 12:08 jcsp

This ran 200x locally without failing. In CI we've seen the original failure very rarely, so it may be a while before any of these trip.

jcsp avatar Aug 09 '22 12:08 jcsp

Hold on. Those ci failures are stranger than I thought. Looking at some briefly

NyaliaLui avatar Aug 11 '22 17:08 NyaliaLui

Those failures are highlighting a nasty behavior where VerifiableProducer can be instantiated on two nodes and their outputs will fight for updates to last_acked_offsets. That's a bug, fixed.

Unfortunately the tests that were failing mysteriously were not the ones with two producers, so the issue these assertions are hunting for is still out there.

jcsp avatar Aug 12 '22 11:08 jcsp

It looks like some of the asserts are triggering.

@jcsp What are next steps here?

NyaliaLui avatar Aug 12 '22 17:08 NyaliaLui

Those last failures were because the logic for whether to use strict validation or not (single or multiple producers) was flipped. Fixed.

jcsp avatar Aug 15 '22 09:08 jcsp

@NyaliaLui please could you re-review + clear your :red_circle: if all good

jcsp avatar Aug 15 '22 20:08 jcsp