search

dominictarr / scuttlebutt

peer-to-peer replicatable data structure
Other
1.32k stars 66 forks source link

Slave mode (non-duplex) #21

Closed refset closed 11 years ago

refset commented 11 years ago

This would be a useful feature for an implicit authorisation model.

dominictarr commented 11 years ago

you can't be a slave with out a master, so you'll need that too.

basically, a master will not allow any update, except from itself https://github.com/dominictarr/scuttlebutt/blob/master/index.js#L67

(or maybe a small set of co-masters)

And a slave should regect local updates https://github.com/dominictarr/scuttlebutt/blob/master/index.js#L59

Also, don't do this if it's a slave https://github.com/dominictarr/scuttlebutt/blob/master/index.js#L147

And skip this if it's in master mode: https://github.com/dominictarr/scuttlebutt/blob/master/index.js#L163

refset commented 11 years ago

Ah, then maybe I don't mean "slave" in the normal sense then, because I want local updates - sorry! Basically I want one-way replication. So you can still use both scuttlebutts like normal.

Master-slave as your describing though would be good for some type of clustering load distribution, I'm sure.

juliangruber commented 11 years ago

You can also check out https://github.com/juliangruber/authoritee which lets you add authoritative nodes to your meshes.

dominictarr commented 11 years ago

@jez0990 what is your use-case?

refset commented 11 years ago

As it's topical: using a decentralised public data set (e.g. npm 2.0) where you don't want private/local data (modules) propagating upstream but keeping a uniform/transparent/minimal API for the layer above (e.g. npm cli)

dominictarr commented 11 years ago

I am concerned about mixing a read only replication with a datamodel you can update. seems like you could get into messy states... are you sure that is the best way?

refset commented 11 years ago

That concern is valid, this would only be appropriate for a class of CRDT models that don't abstract over the history. So crdt's Seq wouldn't work but Set and my DAGDoc thing would.

refset commented 11 years ago

Okay so I just realised that being able to work out retrospectively what is and isn't local/private would be very difficult with what I'm proposing. And that feels important. So scrap this!

I think the solution to what I want is to have two scuttlebutts where contents from one can reference the id's from within the other. No replication between them :)

dominictarr commented 11 years ago

sounds like you are using one scuttlebutt as a local data model - nothing wrong with that, really. you could use that for persistance or, for replicating data across tabs.