Does every node in the network have to know about all transactions issued by other nodes eventually?

Question

With every node I mean the nodes issuing and validating transactions, so certainly full nodes, with light nodes I'm not sure.

I understand that the tangle can be pruned and snapshots can be created so a new node does only have to know all transactions starting from the snapshot so far. I also understand that a node can validate tips without seeing all the transactions that might have been issued somewhere else just a few moments ago. Furthermore, I heard that clusters of nodes can be somehow detached from the network and sync with the tangle later, but then they have to sync and download new transactions later (David Sønstebø talks about it in this interview, listen for 3 minutes for the context or jump directly to 29:00).

So the question remains: Does a node need to see all valid transactions eventually? It appears to me this is the case (excluding invalid transactions not being relayed). Is this correct?

Answer 1

As of the current implementation state, every node will have to receive every confirmed transaction eventually (so that the solid subtangle milestone can increase and the node can calculate all balances if asked for them). There are plans for "sharding" so that nodes only care about certain transactions (e. g. based on first letter of address), but there is no exact specification how they should work later yet.