AutoNAT junk candidates

Open
#7,798 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
3/5
Estimated time
1-2 days
Newbie friendliness
56/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
rust
Domain
networking

Research direction

Start in src/libp2p/discovery.rs and find where Identify candidates enter DiscoveryBehaviour::poll. Use libp2p's swarm address-translation helper to map each observed address to the observed IP with the node's listen port for the same transport, then add a unit test covering that translation. Done means the candidate set stays bounded, the translation test passes, and the 25-server reproduction reaches Public status.

Written by the indexing model from the issue text.

Description

Type: Task

[!CAUTION]
Issue AI-generated while investigating connectivity/reachability issues with crawlers.

Summary

Forest nodes behind a NAT that rewrites outgoing ports (Kubernetes, many Docker setups, CGNAT) never confirm their external address, even when their listen port is forwarded and reachable. They stay AutoNAT Private, don't advertise a public address, and are therefore absent from Lotus routing tables and from network measurements (see #7527).

Cause:

  • Identify reports one observed address per connection. Behind such a NAT each outgoing connection is observed with a different port, so every observation is a distinct, undialable candidate. A home node collected 12 of them within seconds of startup.
  • libp2p-autonat (v1, 0.15) only ever adds candidates and never removes them (behaviour.rs:290). Every probe sends all candidates first, followed by our listen addresses (as_client.rs:190-195).
  • The go-libp2p AutoNAT server that Lotus runs dials the observed address plus at most 15 of the requested addresses, substituting the observed IP (p2p/host/autonat/svc.go:150-193, maxPeerAddresses = 16). Once 15 or more junk candidates have accumulated, observed IP + listen port, the only address that would work, is never tried.

Reproduced locally with a calibnet node that was reachable throughout, connected through a port-rewriting proxy to go-libp2p's AutoNAT server code (only change: loopback dial-backs allowed, so the test could run locally):

AutoNAT servers (junk candidates) Probes Status
5 4/4 succeeded Public, listen address confirmed
25 3/3 failed Private

Proposed fix: identify's candidates pass through DiscoveryBehaviour::poll before they reach AutoNAT. Replace each one with observed IP + our listen port for the same transport (libp2p::swarm::_address_translation). The candidate set then stays small and contains the address that actually works for port-forwarded setups (k8s hostPort, Docker -p, cloud 1:1 NAT).

Completion Criteria

  • Translate external address candidates to observed IP + listen port in src/libp2p/discovery.rs
  • Unit test for the translation
  • Verify with the 25-server reproduction: the node reaches Public
  • Report the unbounded candidate set upstream (rust-libp2p autonat v1)

Additional Links & Resources

  • #7527
  • libp2p-autonat 0.15 src/v1/behaviour.rs:290, src/v1/behaviour/as_client.rs:190-195
  • go-libp2p v0.50.0 p2p/host/autonat/svc.go:125-193
Dominant language
Rust
Stars
700
Forks
201
Avg merge
1d 17h
Merged PRs (30d)
65

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from ChainSafe/forest

All issues in ChainSafe/forest

Similar issues

More Rust issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.