Open adestis-mb opened 1 week ago
I think defining link-tracking for the l3leaf on the child l2leaf would be a bit hard to follow. Would it be good enough if implemented in the l3leaf vars to apply link tracking on all downlinks? I mean the link tracking group used would already be set in the l3leaf vars, so assigning that for downlinks in the same place seems logical to me. This would not allow for selective enabling of link-tracking to certain downlinks, so if that is needed, we will need to combine the ideas somehow.
Let me know what you think and if you need the selective enabling of this.
For us it would be enough to define on the l3leaf what interfaces are downlink interfaces. Normally they are all. It could be that some interfaces should be excluded but this could then be done via endpoints / ports. If the definition in the endpoints / ports takes precedence over the default definition than that should be the best solution.
We are using just one group of uplink (called UPLINK). I cannot say s.th. about others which maybe require more than one group of uplink.
Enhancement summary
In the node definition of a l2leaf (or custom node with l2 services) there must be the possibility to define the link tracking behavior seen from the uplink switch in order to disable the connection if the uplink switch has a core isolation.
Which component of AVD is impacted
eos_designs
Use case example
When you have a pair of l3leafs where an another switch is connected (as l2 network service) as part of the fabric (e.g. via LACP MH), the connection from the l3leaf switch to the l2leaf switch does not go down when there is a core isolation on the l3leaf switch. In case of a core isolation of the parent switch the l2switch would still send traffic which would be dropped on the parent switch.
Describe the solution you would like
In the node definition of the child switch the link_tracking mechanism of the parent switch must be configurable to tell the parent that the connection to the client switch is a downstream connection and the parent should monitor a given uplink connection.
Describe alternatives you have considered
With AVD 4.10.2 the only workaround we found was to add a network port configuration in the connected endpoints where a raw_eos_cli was added:
raw_eos_cli: link tracking group UPLINK downstream
So the workaround looked like the following:
The problem with this solution is that the (ethernet) connection itself is down. Bringing the link up costs a lot of time compared with bringing up the lacp channel.
Additional context
No response
Contributing Guide