For example, if you want to create a new network,
you need to create it on every single nodes in pending state first
is this intentional guideline or can we create something clusterwide
without going through individual nodes first
Correct, this behavior is by design, because we require all nodes to have identical LXD defined networks.
Each member of the cluster can however tweak some configuration details of its own network (such as the “bridge.external_interfaces” key if they have different external bridge interface names).
For this reason creating the network clusterwide in one shot would be a bit problematic because you’d need a way to say “I want this external bridge on this member and that other bridge on that member”, all in a single command line.