Hi there,

I just made my 2nd attempt to upgrade from GL 1.3.4/ES 1.7.1 to GL2.1.1/ES 
2.3.5 and expierenced similar problems
like I had when upgrading to GL2.0 (and decided to rollback).

Issues are mainly in throughput rate - the processing buffer is always at 
100% and the disk journal just keeps on 
growing since more messages come in than can be processed. The top right 
counter often remains
at 0 outgoing messages and peaks sporadically at 2k m/s. In 1.3.4 it 
remains fairly constantly at 2.5k m/s
and the processing buffer fills up at burst times to like 50% maximum. 
Incoming messages peak at 8k
at times.

My setup is fairly straight forward:
2 VMs - 1 VM with Graylog (6 CPU, 12GB RAM), 1 VM with ES-DB (4 CPU, 34GB 
RAM). 
We give ES 18GB RAM (well, Java that is).
Storage wise I keep rotating indexes by size every 40GB and keep 7 indexes 
(roughly 5 days of data)

Configs:
Graylog
1.3.4 (working conf): http://pastebin.com/TjUbNJps
2.1.1: http://pastebin.com/8WYgGYWB

ES
1.7.1: http://pastebin.com/bNh3FY1k
2.3.5: http://pastebin.com/eWm5x4zH

Now, last time I tried upgrading I got answers in IRC that there´s a known 
perfomance bottleneck
with GL not using HTTP as a means of putting data into ES and would likely 
be the issue
I expierenced.

Lastly, this is the log form GL 2.1.1 which had some errors which "might" 
be related just to the
front end (according to 
this: https://github.com/Graylog2/graylog2-server/issues/2345), also, I 
noted
a different TCP-Connection pattern after upgrading: http://imgur.com/n7l8C1b

GL logs: http://pastebin.com/X5K3ZucP

Any help will be appreciated since I´d really like to upgrade to GL 2.x.

-- 
You received this message because you are subscribed to the Google Groups 
"Graylog Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/graylog2/65a972cb-a79f-4e4a-a043-ee90e2734678%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to