feat(hole-punch): add hole-punch interoperability test suite

[thomaseizinger]

How to review this

1. Read the README
2. Look at src/generator.ts
3. Look at testplans.ts

Most of the other files like helpers/cache.ts are a 1-to-1 copy from multidim-interop.

Follow-ups

• Investigate why Rust test has double the RTT for the hole-punched connection on TCP in one direction
• Assert RTT once above is fixed

[thomaseizinger]

Don't be fooled by green CI, it is actually failing :D (But was working locally)

[thomaseizinger]

Don't be fooled by green CI, it is actually failing :D (But was working locally)

Now we actually have passing tests!

[thomaseizinger]

@marten-seemann This is ready to have a go-peer added! I am curious to learn whether the go-peer also observed a weird RTT via the TCP connection. That would tell us whether this is an issue with our network setup or a bug in rust-libp2p somewhere.

[mxinden]

🎉

➜  hole-punch-interop git:(feat/hole-punch-tests) npm run test

> libp2p hole-punch [email protected] test
> ts-node src/*.test.ts && ts-node testplans.ts

Running 2 tests
Running test spec: rust-v0.52 x rust-v0.52 (quic)
Running test spec: rust-v0.52 x rust-v0.52 (tcp)
Expected RTT of direct connection to be ~200ms but was 401ms
0 failures:
Run complete

The amount of code duplication to the existing interop tests is unfortunate. That said, the duplication might just be easier here than a bad abstraction.

[thomaseizinger]

The amount of code duplication to the existing interop tests is unfortunate. That said, the duplication might just be easier here than a bad abstraction.

I have two ideas to deal with this:

1. Get rid of most of the code which just deals with caching: Re-consider container caching strategy to use registries #303
   • Makefiles
   • Cache helpers
   • Build shell-script
   • Version resolution
2. Rewrite the rest to use deno:
   • Deno is much more a "convention over configuration" approach. It directly supports typescript for example.
   • Sharing utility functions across the various test runners should be easy as a result

[thomaseizinger]

@Menduist Would you have time to write a nim-libp2p client for this test suite so we can get a rough idea of whether it is working across implementations at this point?

[thomaseizinger]

Hey everyone! This has been ready for additions of other languages for a while. Instead of having it sit here, my suggestion would be to merge it as is and activate it in rust-libp2p for now. That way, we get some insight on how stable the tests are once they are run on each PR.

Other languages are then free to add their implementation at their own pace.

Any objections @mxinden @marten-seemann @MarcoPolo @achingbrain @maschad @Menduist (not sure who else to tag?)?

[mxinden]

I am in favor of moving forward here. What do others think?

[Menduist]

Sorry, just seen the message, didn't realize this was already ready for integration
@diegomrsantos will add a nim implementation, but feel free to merge in the meantime :)

[thomaseizinger]

I am in favor of moving forward here. What do others think?

I am planning to merge this in ~24h unless there are any concerns raised!