I've got 4 switches connected in a circle and they all run RSTP. Lets assume the following:
switch A (root bridge) ------- switch D | | | | <= that link is the alternate path | | switch B --------------------- switch C
Now to avoid a L2-loop, one link between two switches will be in a blocking state, that port would have the alternate role on one side (that link is between switch C and D). Now when I disconnect the link between A and B, the link between C and D immediately becomes "active" and I've got what I wannted - in the last test I had 2 ping losses, which is ok I'd say. BUT: when I re-establish the link between A and B again, my network goes down for approx 30-40 sec and I can't access devices on switch B from switch A (sounds like traditional STP, but RSTP is configured).
So my question now points to the mac-address-tables: Having a fast convergence (without timers) according to RSTP and avoiding loops (and therefore broadcast storms and mac-address-flapping) is one thing, updating the mac-address-table is another thing. I'd say it isn't enough to enable the link between A and B and to put the link between C and D into the discarding state, because switch A and D need to update their mac address table to make sure that the frames are sent into the anti-clockwise-direction instead of sending frames into the direction of switch C when accessing devices connected to port B.
I've googled a lot, but I have never found any hints or topics regarding the RSTP synchronization in combination with the the mac-address-table update. So the real question is: Are mac-address-tables upgraded when there are topology changes in RSTP or is that not part of RSTP?
EDIT: I've found the issue here, it was a bug in the firmware. Everything's working perfectly now. Thanks!