IPFS eventual consistency on shared folders

[ajres]

When using IPFS for usecase 2:

"As a mounted personal sync folder that automatically versions, publishes, and backs up any writes."

How are concurrent edits on different nodes merged so that the all nodes eventually see the same file system state.

If I understand correctly, each node would write a new commit object on changing the file system, but how are (potentially) conflicting commits on different nodes merged?

Andy

[lucamartinetti]

Very good question. Can somebody please answer?

[fiatjaf]

answer attempt:

IPFS deals with files and hashes of files. There's no concept of "tracking" a file, if a file (or directory) changes and is updated on IPFS, it will be referenced by a different hash. It's up to you, or an external daemon, to keep track of which hash, from those published on IPFS, corresponds to the "newest", or "correct" version of a file.

[mappum]

Conflicting commits can be prevented by using CRDTs.

There's no concept of "tracking" a file, if a file (or directory) changes and is updated on IPFS, it will be referenced by a different hash. It's up to you, or an external daemon, to keep track of which hash, from those published on IPFS, corresponds to the "newest", or "correct" version of a file.

This is handled by IPNS. The path /ipns/HASH_OF_YOUR_PUBKEY will point to the latest version of your directory, since you can sign the latest version's hash and publish it to the DHT.

[jbenet]

To add to this:

• yes, we will handle this (likely with commit objects, or other versioning structures)
• commit objects are WIP -- Versioning: Commit + Repo Datastructures #23
• yes IPNS right now publishes the latest snapshot of a dir, but yes should point to a commit (or other versioning datastructure, such as a CRDT)

[danimesq]

@jbenet
Any news?

[Stebalien]

• We've been doing a lot of work on CRDTs. Check out peerpad.
• The peergos people have been working on building something like this on-top-of IPFS.

IMO, the primary missing component is private content and/or access controls. Right now, that can be layered on-top-of IPFS but that basically means you need to layer a second encrypted filesystem on-top-of IPFS (which is suboptimal...).