Collector group evaluation criteria? [forqandawebinar]
I read in the documentation for v6.0 that agents configured to connect to one of the Collectors within a group are equally distributed within the group. But what is the basis for that distribution? Is this strictly based on count of attached agents per Collector, or (preferably) on a more sophisticated set of criteria, such total purepaths or total events over a certain period of time? Do administrators have some means of controlling the weight of those various factors that come into play in this decision?
Also, the doc goes on to say that if a Collector in a group terminates, the connected Agents will reconnect to another Collector within the same group. But if that Collector group has a total of three or more Collectors, on what basis does the agent choose the Collector to which to connect?
Thanks...
-- Mark
Hi Mark,
In the current iteration (6.0) the collector group feature is handled by randomization. A secondary impact of this feature, though, is that collectors share their instrumentation metadata, so it is now possible to put a real load balancer in front of your collectors and allow agents to failover without missing transaction context data. In this way, you can control how the agents are load balanced to your collectors 
Hope that helps,
Rick B
Richard,
Since you mention you can now put the collectors behind a load balance does that mean I can define a collector group across different servers? And if yes....could one be say Windows and one a flavor of Unix?
Currently our agent point to FQDNs of the servers a particular collector is on. We would like to create a dns entry for a vip and point the agents to that if we can.
Hi John: Yep! Just make sure to put all the collectors behind the same vip in the collector group, again so they share metadata and can translate events from all connected agents no matter who handled the initial instrumentation. Make sense?
Rick B
Rick,
Yes it does. I may have more questions when we get closer to attempting in our test environment but for now I have 1.
- We will probably have collectors of the same name and ports on different servers defined to the same group...Is that OK. The use the LB.
If you put these behind a load balancer and they are in the same collector group, do you need to anything to disable the collector level balancing? Here is what we fear will happen:
Example:
Agent connects to the vip, which passes it onto collector1
Collector 1 responds "you need to reconnect to collector2"
Agent is now connected to collector2 directly, bypassing the vip
Great question: even though rick answered the question I will bring it up today in the webinar
