Using DNS-based load balancing doesn’t save you from the failure. DNS server doesn’t know if the backend is up or down. It just responds without carrying about the state of the backend.
In my previous lighting blog post, I described how it’s possible to null route the whole datacenter by having fat-finger syndrome.
If you use Clos topology you should be aware that there are no direct links between leaves or spines. So, if you announce the default route unconditionally, you know what happens - traffic destined to
::/0) is blackholed. And what happens if you want to announce the default route conditionally? Let’s say I need to announce
only if an arbitrary community
123:1 exists. For instance:
As usual, the latest update broke OSX again. In most cases there are few options to fix this piece of crap:
We had an issue where the whole IPv6 block /32 was blackholed due to fat-fingers syndrome. The mask for the prefix was accidentally missed and IPv6 traffic just went off.
Looks like everyone is involved in this bazaar. Cloud providers should clap hands to increase profit by 30%, Intel, in turn, will sell more chips, increased GDP, etc.
Suppose we have such a local FDB table and wanna to redistribute it via BGP to remote endpoints. We encapsulate this metadata information directly in large BGP community which if large enough to carry all the necessary data - MAC, VLAN, VNI.