Sorry - some final details on this for today. I stopped the broker being
added to the cluster (the one with GC taking 1min & both cores), dropped
the max heap to 1GB (out of 4GB) and started it up again. It joined nicely,
caught up and went ISR. Haven't yet run into the increasing GC cycles
(which
A few more details before the looming crash. The broker replicating data
and in sync (before GC cycle starts) - GC cycles taking nearly 1min real:
2015-03-30T21:08:39.502+0300: 3805.931: [GC2015-03-30T21:08:39.502+0300:
3805.932: [ParNew: 142199K->6219K(153344K), 51.6581120 secs]
678471K->543078K(
Hi,
A related issue which results in a similar increasing GC cycles (but
doesn't recover)... We're trying to bootstrap a new broker into the cluster
(replacing one of the 3 main ones described previously). It connects fine,
starts replicating data and becomes ISR, but starts a similar GC cycle on
Hi,
We are seeing really weird GC behaviour on our Kafka cluster. After a clean
stop/start of the whole cluster, the servers go into a loop of increasingly
slow GC cycles for an hour and then seem to recover to a normal state
without any external changes to the servers or traffic. The load to the
