Looking at some recent JIRAs, such as KAFKA-6568, which came in after the release of 0.11.0
Would that possibly be related to what you observed ? Cheers On Mon, Jul 23, 2018 at 6:23 PM Mitch Seymour <mitchseym...@gmail.com> wrote: > Hi all, > > We're using version 0.11.0 of Kafka (broker and client), and our Kafka > Streams app uses a compacted topic for storing it's state. Here's the > output of kafka-topics.sh --describe: > > Topic:mytopic > PartitionCount:32 > ReplicationFactor:2 > Configs:retention.ms=432000000,cleanup.policy=compact > > The app will write tombstones to this topic when it's finished with a > certain key. I can see the tombstones using kafkacat > > kafkacat -q -b ... -t mytopic -o beginning -c 2 -f"Time: %T, Key: %k, > Message: %s\n" -Z > > Output: > > Time: 1530559667357, Key: key1, Message: NULL > Time: 1530559667466, Key: key2, Message: NULL > > Note: the -Z flag in kafkacat prints null values as NULL to make it easier > to see the tombstones. Anyways, the timestamps on these topics are from > GMT: Monday, July 2, 2018, so I'm not sure why the tombstones still exist > in this topic. > > Furthermore, it looks like compaction is being triggered because I'm seeing > this in the logs: > > discarding tombstones prior to Thu Jul 19 14:15:17 GMT 2018. > > However, I still see tombstones in this topic from Jul 2, so it doesn't add > up Another side note: I'm not explicitly setting delete.retention.ms, and > since the default value is 86400000, or 1 day, I'm not too sure why the > tombstones are sticking around. > > Anyways, has anyone experienced this before? I'm not sure if this is a > known bug, or if there's something peculiar with our own setup. Thanks, > > Mitch >