This is something I have struggled with for a while. I understand the concept of a prober pool, but for anyone who has closely watched the monitoring functioning will find that the GTM seems to use the proper pool settings as more of a recommendation or something. I say this because on several occasions I have found the GTM using a prober NOT within the proper pool to monitor a server. Then I am left trying to figure out why that prober was used. It seems like this is more of a recommendation to GTM, but in the end it is not absolutely followed every time. That doesn't really work well in scenarios where certain probers do not have access to some servers. Anyone have any ideas on what is happening here and maybe how to make it more deterministic?
We're unable to put our brand new GTMs (188.8.131.52) into production because they refuse to honor the prober pool settings and are marking virtual servers down constantly. I've tried setting the configuration both at the data center and server level, tried setting static pools and also the "inside" and "outside" options.
The iQuery algorithm is complex and i have also seen what you describe in practise. It assumes that everything can get to everything but sometimes you want to nail it down to specific devices. I know that in v13 this has been changed to make it more of a recommendation than a hard rule. I guess the only workaround is to ensure that all devices can get to all devices.
Could you provide some documentation that supports this recommendation vs hard rule?
FYI, as part of our support case we were asked to restart gtmd which cleared up our problems.
You'll see that this is changed from Prober Pool to Prober Preference and Prober Fallback. Hopefully this means that they have been doing some work on making this more deterministic as it is currently a black box.