I'm setting up a simple hub-spoke configuration of VNets in Azure. Hub contains a VPN gateway then creating spokes for storage, compute etc. My main concern is latency so I'd rather not setup custom routing through the hub VNet.
Since the storage spoke will need to see the compute spoke, do I need to setup peering connections between the spokes or does Azure route through the hub VNet automatically?
Without the routing through an NVA (which also requires you to enable transit explicitly on the VNet), VNet traffic is not transitive.
If you have two spokes peered to an hub, they don't automatically get to talk to each other.
You need to set up peering between the two spokes, too.
It is practical when you have few spokes, but it does not scale if you mean to add more as the number of peerings will grow exponentially.