Sounds good, let us know how it goes. There are many tuning options with ES and in Metron, a lot of which are described in the guide I put together, but some are definitely missing. That said, the guide could use an update and if you can think of anything that would help make it more clear I'd be happy to clean it up. I also have on my to do list to move it into the GitHub code, and thus the generated site-book(s).
Jon On Wed, Mar 29, 2017, 1:06 PM tkg_cangkul <yuza.ras...@gmail.com> wrote: > hmm ok. > > so it means i better run this metron on separate cluster with my other > apps right? > > ok thank you so much for all of your help guys. :) > i'll try it soon after i migrate my other apps. > > > > On 29/03/17 23:52, Simon Elliston Ball wrote: > > In general for any ES use case you should be running only ES on the ES > nodes, and yes, you should use min(31, RAM / 2) as heap size. It’s also > worth considering shard counts for indices and disk layout in the ES nodes. > > As Nick points out, you’re also going to find it hard to tune ES for large > numbers of applications running different indexing, so it’s generally a > good idea to build different ES environments for your multiple use cases. > > Simon > > On 29 Mar 2017, at 17:02, tkg_cangkul <yuza.ras...@gmail.com> wrote: > > Well, > > actually there are not only metron that running on my cluster. there are > some other apps running on it. i'm afraid it will be crash if i set it to > 50% of my RAM. Are ES heap ideally set to 50% RAM memory to get the maximum > performance? > > > On 29/03/17 17:18, zeo...@gmail.com wrote: > > Right off the bat I would give 31GB heap to each ES node. Normally you > want that number close to 32GB but not exceeding 50% of your total server > RAM so it can be used for disk caching. Let me know if that helps, I'd be > happy to help further. > > Jon > > On Wed, Mar 29, 2017, 12:56 AM tkg_cangkul <yuza.ras...@gmail.com> wrote: > > hi Zeolla, > > thanks before for your reply. > sorry if i changed the email subject. i think the problem of kafka topic > partition has been solved. but the indexing still slow for me. > > this is my cluster environtment for ES: > i have 5 servers ( 1 master, 4 datanode ) > every server there is 64Gb memory on each machine. > > i have set 4Gb to my ES Heap Size. 4 Partition of kafka topic but the > indexing still slow. > > any suggest to tuning the indexing? > > > On 28/03/17 08:43, zeo...@gmail.com wrote: > > Can you clarify what you mean by recreate your kafka topics? Usually what > I do to add partitions to a kafka topic in Metron is something like: > > zk=server1:2181;/usr/hdp/2.5.0.0-1245/kafka/bin/kafka-topics.sh > --zookeeper $zk --alter --topic bro --partitions 4 > > Once you run this, your bro topic (or whatever you specify next to topic) > will now have multiple partitions. It is effectively transparent to the > rest of Metron. > > That said, are you running kafka and ES on the same boxes? You say you > want 4 partitions because you have 4 ES boxes, but it should really map to > the # of boxes/disks you have running kafka brokers. Would you mind > providing a basic layout of your environment (what servers run where, how > many servers, etc.)? Thanks, > > Jon > > On Fri, Mar 24, 2017 at 11:56 AM tkg_cangkul <yuza.ras...@gmail.com> > wrote: > > anyone can help me for solving this problem? > > > On 24/03/17 18:24, tkg_cangkul wrote: > > hi, > > i wanna try to tuning my ES when indexing data from all the sensors (bro, > yaf, snort). > i've read this article : > > https://cwiki.apache.org/confluence/display/METRON/Tuning+the+Search+tier > > on point 3. Assuming that ES will get behind from time to time, you may > want to increase the indexing kafka topic size. You should also consider > increasing the partitioning kafka topics in general to distribute the load > better and increase parallelism. > > then i try to recreate my kafka topic to 4 partition because i have 4 > datanode on my elastic search. then, i check the broker node in zookeeper > for all the topics that i create and i've seen there are *[1,2,3]* > partition inside the */brokers/topics/(all_my_topics_name)/partitions* > But when i start the storm topology, i've seen this error message on my > storm logs if there is no node for > > */brokers/topics/(all_my_topics_name)/partitions/0/state *Why it still > read the *0 *partitions.? is the metron just set for assigned 1 partition > only on kafka topic? > then i try to create manually the > */brokers/topics/(all_my_topics_name)/partitions/0/state > *but i've found another error msg like this: > > > > any suggestion about this? > > > > > -- > > Jon > > > -- > > Jon > > > > > -- Jon
