fix: don't add unresponsive DHT servers to the Routing Table

[guillaumemichel]

PR addressing #811

Change summary:

• added a lookupCheck function to the IpfsDHT. This function consists in making a DHT FindPeer request to a peer, asking it for its own peer.ID https://github.com/guillaumemichel/go-libp2p-kad-dht/blob/cdd6898b8a9d41a1abce803325c98a14083013d6/dht.go#L327-L336
• added a validPeerFound function, acting like the previous peerFound function, but assuming the given peer was just queried https://github.com/guillaumemichel/go-libp2p-kad-dht/blob/cdd6898b8a9d41a1abce803325c98a14083013d6/dht.go#L664-L676
• peerFound performs the validRTPeer check and runs protocolCheck. Upon success it calls validPeerFound https://github.com/guillaumemichel/go-libp2p-kad-dht/blob/cdd6898b8a9d41a1abce803325c98a14083013d6/dht.go#L639-L662
• peers that were just successfully queried with a DHT lookup are directly sent to validPeerFound https://github.com/guillaumemichel/go-libp2p-kad-dht/blob/cdd6898b8a9d41a1abce803325c98a14083013d6/query.go#L419-L420
• IpfsDHT has a new recentlyCheckedPeers attribute, a map of the peerids contacted in the last lookupCheckInterval (5 seconds by default). It prevents network loops, e.g A sends a request to B, B wants to add A in its RT, B sends a request to A before adding it to its RT, A gets the requests and want to add B in its RT so it sends a new request to B, etc. If we tried to add a peer to the RT less than 5 seconds ago, we won't try again.
• Before A sends a lookup request to check whether B behaves as expected, check if there is available space in A's RT in the bucket where B belongs. If this bucket is full, abort adding B to the RT and sending any request.

Todo

• Implementation
• Reviews and approval
• Merge UsefulPeer function go-libp2p-kbucket#113
• Update dependencies to https://github.com/libp2p/go-libp2p-kbucket
• Update relevant documentation
• Merge

[guillaumemichel]

@Jorropo I fixed a flaw but some tests are still failing:

• TestBootstrappersReplacable doesn't make sense to me. The concept of replaceable peers seems wrongly implemented in go-libp2p-kbucket. I suggest that we get rid of this test.
• All tests from ext_test.go are failing. They are testing whether DHT requests fail as expected in scenarios the DHT peer is connected to unresponsive peers. However, with this new feature, invalid nodes aren't added to the routing table at all, hence the tests are failing.
• All other tests are passing as expected.

For ext_test.go, I suggest to get rid of all tests, and add a new test where we try to connect to unresponsive peers and verify that they don't get added to the routing table. WDYT?

[guillaumemichel]

Just opened a PR in go-libp2p-kbucket. We need a function to test whether a peer.ID would be useful to the Routing Table, so that we don't bother sending requests to useless peers.

libp2p/go-libp2p-kbucket#113

[Jorropo]

I did a high level surface review, I'll rereview in depth once high level code quality is fixed.

[yiannisbot]

@guillaumemichel is this ready for a second round of reviews from @Jorropo so that we schedule it for one of the upcoming releases?
cc: @BigLep to be in the loop regarding scheduling this, or deferring for later depending on team's priorities.

[BigLep]

@guillaumemichel : what are the next steps here?

[guillaumemichel]

I am still waiting on a review from @Jorropo for this PR and for libp2p/go-libp2p-kbucket#113 on which it depends.

[Jorropo]

Also please resolve your conflicts.

[BigLep]

Per 2023-06-08 maintainer conversation: @Jorropo is going to take having this code all ready to go so that @guillaumemichel can review it first-thing 2023-06-12 when he returns so that the Kubo 0.21 release can happen afterwards.

[Jorropo]

LGTM Ty <3