provide initial tokens

[rustyrazorblade]

trafficstars

We currently start the cluster without worrying about tokens allocation or consistent range movement.

It would be a lot better if we could calculate the tokens for each machine before pushing it up. We might need to abandon parallel-rsync for this and instead create one Kotlin coroutine for each server, which would allow us to edit the yaml file on the fly before pushing it up.

[jsanda]

@rustyrazorblade would this involve doing something like what is described in setting up a cluster with even token distribution?

[rustyrazorblade]

I thought about it, but what we should really be doing is providing initial_token: rather than relying on allocate tokens for keyspace. initial token will work with all versions as well as with a cluster where we fire up 50 nodes all at once. Using allocate tokens for keyspace means we have to be a little more careful and take a lot longer.