I'm grasping at straws but maybe play with your batch sizes? I hear now there's a time based option too that I haven't played with.
It would be helpful to have more metrics about resource utilization over time in your cluster, are you able to gather anything and maybe put it into a visualization tool? Jon On Thu, Aug 17, 2017, 16:35 Laurens Vets <laur...@daemon.be> wrote: > That seems close to the issue that I'm having... But there's a part I > don't quite understand in my case. > > Everything's working fine for days and then suddenly, Java throws an > error (see other mail to the mailing list) and I can't get Metron stable > again. It's always either the indexingBolt or hdfsIndexingBolt. The > error I see the most after this is "java.lang.OutOfMemoryError: GC > overhead limit exceeded"... > > It seems that somewhere there's a big pileup of messages which Metron > suddenly can't process. > > Any ideas on how to further troubleshoot this? > > On 2017-08-17 11:10, zeo...@gmail.com wrote: > > I used to run into similar issues when my environment was resource > > constrained but never ran it to root cause. It has been a long time > > since > > I was in this scenario to re-test. > > > > https://issues.apache.org/jira/projects/METRON/issues/METRON-485 > > > > Jon > > > > On Thu, Aug 17, 2017 at 12:49 PM Laurens Vets <laur...@daemon.be> > > wrote: > > > >> Hi, > >> > >> Simple question, when I stop Metron, Kafka & Storm via Ambari, I still > >> see the storm worker processes running, is this expected? > >> > -- Jon