More Thorough Autorelay Tests

[ZenGround0]

The filecoin project uses the autorelay and autoNAT features and is aware that it is not trivial to set up tests for these things. The existing autorelay test is reassuring to have. To further improve confidence in the reliability of this feature we would love enhancements to the current autorelay tests. This includes deflaking as well as adding more extensive tests of the features, ideally tests using more realistic networks with less mocking.

cc @phritz -- call out anything I'm missing.

[Stebalien]

See the comments starting at ipfs/kubo#5785 (comment) for some discussion on this topic.

[ghost]

@ZenGround0 Could you verify that this is still a P1 priority for Filecoin? Does it block the launch of testnet or mainnet?

For background: we ran ~16 autorelay servers for IPFS for several months earlier this year. They performed well in the wild. Their bandwidth bills unfortunately proved to be astronomically high, and even after soliciting competitive bids from different hosting companies, we were ultimately forced to all but abandon the project (we have scaled back to either 4 or 2 servers only today). The current focus is on eliminating most of the work relay nodes must do by immediately upgrading relayed connections to direct p2p connections and then removing ourselves from the whole data transfer picture.

We could ask @vyzo to pause his work on that direct connection stuff and build more tests, but my gut feeling is that it's not really what Filecoin wants. Although AutoRelay will be better tested, it will still be too expensive to offer broadly. My recommendation is we try to get a working, well-tested direct connection upgrade version of relay by December 11, and fall back to classic relay nodes only if we fail to hit that date. @vyzo wdyt? @ZenGround0 wdyt? @Stebalien any wisdom here?