Hi Leo,
We also have a fairly idle CPU for our brokers, but we aren't tracking
latency like you are. We are mostly interested in overall throughput and
storage. We are using compression and bundling which add to latency but
decrease storage usage.
Good luck,
Cliff
On Wed, Feb 3, 2016 at 10:55
Hi Cliff,
Thanks for getting back to me on this. Sorry for the delay on replying to
you, but I was away on holiday.
I have been monitoring the Kafka brokers via JMX and on all nodes CPU usage
was pretty much idle, under 1% as you said. I should have mentioned before
that the Kafka cluster was not
We fixed two serious Snappy bugs for 0.8.2.2, so you may want to check that.
On Thu, Jan 21, 2016 at 8:16 AM, Cliff Rhyne wrote:
> Hi Leo,
>
> I'm not sure if this is the issue you're encountering, but this is what we
> found when we went from 0.8.1.1 to 0.8.2.1.
>
> Snappy compression didn't wo
Hi Leo,
I'm not sure if this is the issue you're encountering, but this is what we
found when we went from 0.8.1.1 to 0.8.2.1.
Snappy compression didn't work as expected. Something in the library broke
compressing bundles of messages and each message was compressed
individually (which for us cau
Hi all,
We are using Kafka in production and we have been facing some performance
degradation of the cluster, apparently after the cluster is a bit "old".
We have our production cluster which is up and running since 31/12/2015 and
performance tests on our application measuring a full round trip
