Sooo….I’m playing with MLAG and spanning-tree (RSTP to be exact).
Take a pair of CRS3XX (Pair A) switches and make an MLAG pair, and set the pair to be STP Priority 0x2000
Make a second pair of CRS3XX switches (Pair B) as MLAG pair and set it to STP Priority 0x4000.
VLAN filtering is enabled everywhere, uplink ports are configured to only admit VLAN tagged traffic. The peer-group VLAN is isolated to just the specific pair of switches in the MLAG pair. Only actual data VLANs are allowed to flow north-south.
Now…Plug a link from Pair A, SW1, to Pair B, SW1.
Both switches in Pair A properly see themselves as the root bridge with Priority 0x2000.
Switch 1 in Pair B properly recognizes the root bridge via the uplink connection to Pair A, SW1.
Switch 2 in Pair B thinks that it is THE ROOT BRIDGE with priority 0x4000.
I guess technically, this is correct, but it “feels” off to me. Shouldn’t Pair B, SW2, “see” the STP root path back to Pair A via Pair B SW1?
In a stacked switch (instead of MLAG pair), there would be a single control plane and as such, the “stack” (MLAG Pair B in this case), would be 0x4000 (added clarification: and recognize the root bridge on the uplink port to Pair A SW1).
If I were to add a second link between Pair A SW2, and Pair B SW2, then I would expect that Pair B SW2 would correctly see that it is not the root bridge. Typically that second link would be part of client bond and all of the normal LAG rules would apply. If that second link is of lesser capability than the link between Pair A SW1/Pair B SW1, then I think just adding some port cost would allow it go into an alternate forwarding state between the pairs.
All of these devices are on 7.20.1/7.20.2.