Hi team,
I have a topic with 2 partitions in Kafka. I produced all data to partition
0 and no data to partition 1. I created a Flink job with parallelism to 1
that consumes that topic and count the events with session event window (5
seconds gap). It turned out that the session event window was ne
Hi,
This is expected behaviour due to how the per-partition watermarks are designed
in the Kafka consumer, but I think it’s probably a good idea to handle idle
partitions also when the Kafka consumer itself emits watermarks. I’ve filed a
JIRA issue for this: https://issues.apache.org/jira/brows
The case I described was for experiment only but data skewness would happen
in production. The current implementation will block the watermark emission
to downstream until all partition move forward which has great impact on
latency. It may be a good idea to expose an API to users to decide what th
A possible workaround while waiting for FLINK-5479, if someone is hitting
the same problem: we chose to send "heartbeat" messages periodically to all
topics & partitions found on our Kafka. We do that through the service that
normally writes to our Kafka. This way every partition always has some
~r
I'm also affected by this behavior. There are no updates in FLINK-5479 but
did you manage to find a way to workaround this?
Thanks,
Gerard
--
Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/
Hi,
I've just elevated FLINK-5479 to BLOCKER for 1.5.
Unfortunately, AFAIK there is no easy workaround solution for this issue
yet in the releases so far.
The min watermark logic that controls per-partition watermark emission is
hidden inside the consumer, making it hard to work around it.
One p
Thanks Gordon.
Don't worry, I'll be careful to not have empty partitions until the next
release.
Also, I'll keep an eye to FLINK-5479 and if at some point I see that there
is a fix and the issue bothers us too much I'll try to apply the patch
myself to the latest stable release.
Gerard
On Wed, D