Hello Juniper experts,
I need an advice regarding below testing scenario, maybe someone already did that or tested it in a lab and can give an advice about the best practice to follow.
As in the attached topology we have a VC of 2 spines and 4 leafs, the fail-over test will include:1- Disabling UPlinks ports on S01 checking all is ok then the enabling it and do the same for S02, this quit straightforward.2- Switching master from S01 to S02 and rollback, also obvious.
3- Disabling VC ports towards leafs (not all) but one per leaf every time and make sure leafs are still connecting to the other spine and traffic is going, then enabling it and do the same for next.and here few questions:- Is it better to do that on Leafs side or spine side?- Is it better to do it physically (removing cable) or by commandsOn member: > request virtual-chassis vc-port set interface vcp-0 member 0 disableOn MAster: > request virtual-chassis vc-port set pic-slot 0 port 48
it is production environment so what do you think there could be any related issues I need to take care about or it will be straightforward like disabling any port? what approach to take?
What made me concerned about it is this article:[QFX] Disabling and enabling one VCP port is bringing down the second VCP port on QFX5100https://kb.juniper.net/InfoCenter/index?page=content&id=KB33338&actp=METADATA
The kb does provide a mitigation in the disabling of auto-channelization that should allow you to perform the test.
But another method to consider is doing all these tests as physical ones. This is inconvenient as someone must be on site and pull cables. But I prefer this method because it is the actual failure that will occur, the loss of the physical link.
This also has the benefit of generating the link down alarms that you can use to test your monitoring systems and be sure such events are seen and can also test the alert visibility process too.