Issue with Graylog inputs behind loadbalancer


(Sascha Henke) #1

Hi all,

yesterday I tried to put a hardware loadbalancer (f5) in front of our 2 GL servers for a TCP input.

On the LB I set the virtual server with the correct TCP port (5555) and configured the GL nodes as backend with their corresponding global input on port 5555 with roundrobin.
I could see the open port with netstat and see incoming traffic with tcpdump on the LB.
On one node I could see traffic from the LB to the port (checked with tcpdump), but not on the other node.

On the input in the GL web interface I was able to see incoming traffic, but no messages were indexed.
We used a device which normally logs on a different input on our GL server.

Our setup:
1 GL server (want to put a second node in the cluster to distribute traffic)
3 node ES cluster
3 node MongoDB replica set
All servers are running CentOS7, version is “Graylog 2.5.1+34194da”, ES is still on 5.6.14, installed with RPMs

I hope someone can help us, I was only able to find “how to’s” with the GL web interface behind a LB.

Best,
Sascha


(Jan Doberstein) #2

When you use TCP you need to configure your LB in the correct way - if both nodes are “alive” new incoming connections should be distributed. When the sending client holds the TCP connection the backend will always be the same that gets everything.


(Sascha Henke) #3

Okay, understood that.
But what about not seeing any indexed messages? I can see incoming traffic in the input detail, but when I click on “Show received messages”, nothing is shown.


(Jan Doberstein) #4

is your journal raising or are the message ingested into elasticsearch?

  • check the journal
  • check the Graylog server.log
  • recalculate index range “System > indices > Index name > maintenance > recalculate index range”