Hi Pete, thanks for sharing this!
In general, using real hardware for database applications like Elasticssearch is always a good idea, especially since it prevents problems caused with noisy neighbors and disk-cache thrashing in virtualized environments. Cheers, Jochen On Thursday, 8 January 2015 02:03:39 UTC+1, Pete GS wrote: > > I've recently been through this in another thread here and some very > helpful replies had me looking at ElasticHQ also and this is an excellent > plugin that helped me understand my Elasticsearch nodes were CPU bound. > > I have gone away from VM's for Elasticsearch and am instead now using > physical servers. 4 x dual 6 core blades with Hyperthreading enabled and > 72GB RAM run my "active" Elasticsearch indices for Graylog2. Two additional > dual 4 core blades with 72GB RAM run my "archive" indices which we rarely > access. > > I have also set the refresh interval to 30 seconds as recommended and > since doing this in conjunction with closing older indices everything has > been reliable and stable and Graylog2 is very responsive to searches. > > I also refined buffer processor settings and batch sizes as per > recommendations here, so all these parameters made a huge difference to the > environment. > > My Graylog2 servers are still VM's and working fine (12 CPU's, 32GB RAM). > I did however add two lower spec Graylog2 servers which I use solely for > the web interface, one of which is the Master node. Neither of these have > been added to our load balancer config for the inputs so they are dedicated > search only. > > Ours is now live and happily servicing an average of 3000 to 4000 messages > per second but this morning it had a spike up over 10000 per second and > still worked well. > > Hope this helps... > > On Saturday, 3 January 2015 23:43:18 UTC+10, Joseph DJOMEDA wrote: >> >> Hello guys and happy new year. >> >> I am really sorry for this very noob question. I have a cluster (I am not >> even sure I set up correctly) of 3 nodes. >> >> 1st node: central logstash/graylogs UI/ graylogs Server all installed >> using their ubuntu repositories. >> second and third nodes are solely elasticsearch instances both installed >> with ubuntu repository >> >> kindly find on pastie the configuration <http://pastie.org/9810898> I >> used. >> >> Each box is a VM of : >> >> >> - *2 dedicated cores* >> - *6 virtual cores* >> - *12 GB RAM guaranteed* >> >> >> >> - I thought I would get away with this but I get : >> >> Nodes with too long GC pauses 15 hours ago >> There are Graylog2 nodes on which the garbage collector runs too long. >> Garbage collection runs should be as short as possible. Please check >> whether those nodes are healthy. (Node: *4108686b-xxxxxxxxxb6400*, GC >> duration: *1022 ms*, GC threshold: *1000 ms*) >> >> While memory stabilized at 4GB the CPUs are almost 100% constantly for >> graylogs UI/Server node. From all I read this type of issue is due to lack >> of memory. I had to resume some of the streams because it was automatically >> pause I presume due to the same error. >> >> Please let me know whether my setup is ok (not sure about the cluster bit >> on elasticsearch) >> Please help me fix the GC issue as well. >> >> Thanks >> > -- You received this message because you are subscribed to the Google Groups "graylog2" group. To unsubscribe from this group and stop receiving emails from it, send an email to graylog2+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.