Skip to content

Latest commit

 

History

History
168 lines (124 loc) · 8.99 KB

README.md

File metadata and controls

168 lines (124 loc) · 8.99 KB
Raw

Kitsune: a go-libp2p proxy/load balancer

Kitsune is a proxy/load balancer based on libp2p. It is currently centered on load-balancing IPFS traffic over bitswap.

Why?

Working document

libp2p is a great peer-to-peer network stack. It has awesome features such as built-in encryption, firewall hole-punching and network protocol transparency. However, some of these features come with their own costs.

Specifically, one problem we have experienced at Protocol Labs is that it is very difficult or impossible to horizontally scale a node, especially when clients have hardcoded Multiaddresses to point to specific peers.

One specific case is the js-ipfs preload nodes, which are used by js-ipfs for 2 purposes:

  • As a proxy to the DHT (since js-ipfs in the browser is limited in the number of connections that it can establish, and cannot listen for connections).
  • As a persistence mechanism. Every time content is added to a js-ipfs node, it will also cause the preload node to fetch that content from itself (via their /api/v0/refs endpoint), so that it's persisted even if the page is reloaded.

Since the preload nodes also have to act as bootstrap nodes, their multiaddresses have to be added to the js-ipfs configuration. Since they are bootstrap nodes, their full multiaddresses (including their Node IDs) are nodes are hardcoded in the library itself. This also means that adding a new node means releasing a new version of the library. While we have added preload nodes as load increased, there are applications out there that are still running with outdated versions of the library which know only of a subset of them. This has resulted in increasing load on the first preload nodes that were installed, and the only way to reduce the load is by scaling up. In the long term, this is unsustainable.

The idea behind Kitsune is to replace the preload nodes with a load balancer, which will then balance the load between downstream go-ipfs nodes. The load balancer will take over the identity of the current preload nodes, allowing us to scale them out.

Terminology

  • A downstream node or peer is one of the nodes behind the reverse proxy (e.g. the go-ipfs preload nodes)
  • An upstream node or peer is a node that requests data from the reverse proxy (e.g. the js-ipfs nodes)

TODO diagrams

Project goals

The initial project goal is to be able to replace the preload nodes with Kitsune instances, each backed by multiple go-ipfs nodes. This means that:

  • Not all protocols that IPFS uses will be initially implemented
  • Those protocols which are implemented will only be partially implemented, with enough functionality to work with the preload nodes.

Longer-term I hope to be able to more fully implement the protocols used by IPFS.

A side goal is to serve as a basis for creating debugging and monitoring tools.

Additional uses

p2p-p2p bridge

Kitsune can also be used to bridge between IPFS swarms, allowing for one-way traffic without polluting either swarm's DHT. For example, by implementing support for private upstream swarms (currently unimplemented), you could deploy an instance of Kitsune backed by one or more nodes in the public swarm and peer it with all of the hosts in the private swarm (since it does not participate in the DHT). This would enable the hosts in the private swarm to get data from the public swarm, but not vice versa.

Project status

Kitsune is under development. At this point it has successfully connected 2 go-ipfs instances and transferred files between them using Bitswap, as well as getting a file from a different go-ipfs node by proxying via the downstream node.

Compiling

Just run go build. You will need Go 1.17 (this is tested on 1.17.8). A .tool-versions file is included if you use asdf.

Running

$ GOLOG_LOG_LEVEL=info ./go-libp2p-kitsune -d /ip4/127.0.0.1/tcp/5001 -l /ip4/0.0.0.0/tcp/4441
2022-03-23T14:07:59.036+0100	INFO	main	go-libp2p-kitsune/main.go:111	Peer ID: QmQGKBtcrBLFTfKQNVKMRkfQDvvrPUdR2j3JDKXAN4K9Fg
2022-03-23T14:07:59.036+0100	INFO	main	go-libp2p-kitsune/main.go:112	Proxy addresses:
2022-03-23T14:07:59.036+0100	INFO	main	go-libp2p-kitsune/main.go:210	    /ip4/192.168.1.102/tcp/4441/p2p/QmQGKBtcrBLFTfKQNVKMRkfQDvvrPUdR2j3JDKXAN4K9Fg
2022-03-23T14:07:59.036+0100	INFO	main	go-libp2p-kitsune/main.go:210	    /ip4/127.0.0.1/tcp/4441/p2p/QmQGKBtcrBLFTfKQNVKMRkfQDvvrPUdR2j3JDKXAN4K9Fg
2022-03-23T14:07:59.036+0100	INFO	main	go-libp2p-kitsune/main.go:114	Downstream peers:
2022-03-23T14:07:59.036+0100	INFO	main	go-libp2p-kitsune/main.go:210	    /ip4/127.0.0.1/tcp/4001/p2p/12D3KooWArqy6bE4wH8UnHABYXYtid9uaCpi6kkar8im69KzKjcU

2022-03-23T14:07:59.037+0100	INFO	prometheus	prometheus/prometheus.go:78	Prometheus server started on port 9090

The binary is called go-libp2p-kitsune and accepts the following flags:

  • -d <multiaddr> (mandatory) Comma-separated list of downstream go-ipfs node RPC API port multiaddrs (e.g. /ip4/10.0.1.42/tcp/5001,/ip4/10.0.5.3/tcp/5002). The backing go-ipfs nodes must allow access to the /api/v0/id and /api/v0/refs HTTP RPC endpoints.

  • -l <multiaddr> (optional) Multiaddr to listen on for TCP/UDP traffic (e.g. /ip4/127.0.0.1/tcp/4441). IP 0.0.0.0 means listening on all available IP addresses. The default is /ip4/0.0.0.0/tcp/0 (all IP addresses, random port).

  • -w <multiaddr> (optional in normal mode, mandatory in preload mode) Multiaddr to listen on for WebSocket traffic, with or without the /ws protocol (e.g. /ip4/127.0.0.1/tcp/8442 or /ip4/127.0.0.1/tcp/8442/ws). IP 0.0.0.0 means listening on all available IP addresses. The default is /ip4/0.0.0.0/tcp/0 (all IP addresses, random port).

  • -k <file> (optional) Name of the keyfile. When Kitsune starts up the first time it will store its private key in this file. The next times it will read this file to get its identity/private key. This is the same key format as in ipfs key import|export. The default is ./key

  • -p <port> (optional) Enable preload mode and indicate the port used for exposing the /api/v0/refs endpoint (see below)

Preload mode

> GOLOG_LOG_LEVEL=info ./go-libp2p-kitsune -d /ip4/127.0.0.1/tcp/5001 -l /ip4/0.0.0.0/tcp/4441 -p 8441 -w /ip4/0.0.0.0/tcp/8442
2022-03-23T14:14:58.173+0100	INFO	main	go-libp2p-kitsune/main.go:111	Peer ID: QmQGKBtcrBLFTfKQNVKMRkfQDvvrPUdR2j3JDKXAN4K9Fg
2022-03-23T14:14:58.173+0100	INFO	main	go-libp2p-kitsune/main.go:112	Proxy addresses:
2022-03-23T14:14:58.173+0100	INFO	main	go-libp2p-kitsune/main.go:210	    /ip4/192.168.1.102/tcp/4441/p2p/QmQGKBtcrBLFTfKQNVKMRkfQDvvrPUdR2j3JDKXAN4K9Fg
2022-03-23T14:14:58.174+0100	INFO	main	go-libp2p-kitsune/main.go:210	    /ip4/127.0.0.1/tcp/4441/p2p/QmQGKBtcrBLFTfKQNVKMRkfQDvvrPUdR2j3JDKXAN4K9Fg
2022-03-23T14:14:58.174+0100	INFO	main	go-libp2p-kitsune/main.go:210	    /ip4/192.168.1.102/tcp/8442/ws/p2p/QmQGKBtcrBLFTfKQNVKMRkfQDvvrPUdR2j3JDKXAN4K9Fg
2022-03-23T14:14:58.174+0100	INFO	main	go-libp2p-kitsune/main.go:210	    /ip4/127.0.0.1/tcp/8442/ws/p2p/QmQGKBtcrBLFTfKQNVKMRkfQDvvrPUdR2j3JDKXAN4K9Fg
2022-03-23T14:14:58.174+0100	INFO	main	go-libp2p-kitsune/main.go:114	Downstream peers:
2022-03-23T14:14:58.174+0100	INFO	main	go-libp2p-kitsune/main.go:210	    /ip4/127.0.0.1/tcp/4001/p2p/12D3KooWArqy6bE4wH8UnHABYXYtid9uaCpi6kkar8im69KzKjcU
2022-03-23T14:14:58.174+0100	INFO	main	go-libp2p-kitsune/preload_http.go:17	Preload mode enabled with HTTP /api/v0/refs on port 8441

2022-03-23T14:14:58.174+0100	INFO	prometheus	prometheus/prometheus.go:78	Prometheus server started on port 9090

Normally Kitsune will forward bitswap traffic in one direction only:

  • WANTs go from upstream nodes to downstream nodes
  • BLOCKs go from downstream nodes to upstream nodes

This is enough to e.g. connect 2 go-ipfs nodes (or swarms) and allow the upstream swarm to get blocks from the downstream swarm and not vice versa, without adding the upstream host to the downstream swarm's DHT.

Enabling the preload functionality allows Kitsune to act as a proxy for the js-ipfs preload nodes. Apart from the basic functionality of allowing upstream peers to get bitswap data from the downstream peers, it will enable the /api/v0/refs RPC endpoint. The preloads use this for 2 things:

  • To pre-cache a whole IPLD graph and allow a js-ipfs node to get all the associated blocks in parallel
  • When adding a file to IPFS, js-ipfs will call /api/v0/refs?recursive=true&arg=<CID>. This starts a bitswap session from the preload node to the js-ipfs node, so that the data is preserved in case e.g. of a browser reload

This second case means that we need to enable a limited version of Bitswap going the inverse way: WANTs from downstream nodes and BLOCKs to upstream nodes. Calling /api/v0/refs?recursive=true&arg=<CID> will enable this transfer for the specific CID given and its child CIDs.