What are the Consequences of Pruning Blocks on Validators?

[zicklag]

For my Substrate use-case, I'm running my substrate chain on a cluster of servers all run by a single organization/operator.

I'm using the blockchain as kind of a general-purpose, fault-tolerant database, so being able to trace the current state all the way to genesis may not be a requirement to some users, and I'm a little worried about the block history taking up an unbounded amount of disk space over time.

Is it possible for me to only keep the last so many blocks on all of the nodes in the cluster ( i.e. there isn't a single node running in archive mode ) and still maintain cluster stability as new nodes try to join the cluster?

Can new nodes join by syncing the current state instead of replaying the whole chain?

I was also thinking that it would be nice if I could export the blocks to S3 storage maybe, so that even though I don't have an archive node running, I can still persist the whole chain in an S3 bucket.

Any thoughts or guidance is appreciated!