Storage backend version 2

[jfinkhaeuser]

This might be an IPFS backend, or a file system backend, that's still somewhat undecided. However, we can scope out each approach fairly well:

Filesystem

1. Keep Storage file system layout, but
2. at each leaf don't manage a hyperdrive repository, but only a single content file.
3. Repository implementation for filesystem effectively merges into Storage.
4. Synchronization fetches from other nodes, and stores/deletes on file system.

IPFS

1. Use https://github.com/ipfs/js-ipfs (no DHT yet) OR the HTTP proxy of a standalone node as the backend.
2. Synchronization pins/removes content IDs.

[mnaamani]

HTTP API to talk to an IPFS node, If we have to use the go implementation and run it as a seprate process.

https://docs.ipfs.io/reference/api/http/

client library: https://github.com/ipfs/js-ipfs-http-client

[mnaamani]

We will want to have a "closed IPFS network" without need for a pre-shared secret among the nodes.
This can be achieved by configuring nodes not to bootstrap from global IPFS network. So we will have our own joystream bootstrap nodes, and nodes can use filters to allow access from authorized peers.

[mnaamani]

After discussing over last few days pros and cons of IPFS, I have decided that there are more benefits in going direction of IPFS and I believe it can help us move forward more quickly.