[issue #1250] Proposal to add better announce and/or observed addresses, reachable public ip in host

[asabya]

When a node runs on AWS EC2 or behind a home router, the OS only knows its private IP (e.g. 10.0.0.5). It has no way to learn its public-facing IP. This means the node advertises only private/local addresses to the network, so peers outside the local network can't reach it.

There is no separation between "listen addresses" and "advertised addresses" in py-libp2p.

Proposed solution to display public reachable ip/addresses

• Add announce_addrs as configuration in the host while init, which will be returned in the get_addr() (js-libp2p style)
• Add address_factory as configuration in the host while init, which will manupulate the addrs in the get_addr() (go-libp2p style)
• use UPnP get_external_ip() to inject the external IP into get_addrs()
• Make a ObservedAddrManager that will track what other peers report as host address, and after sufficient confirmations, injects it in get_addrs()

[seetadev]

@asabya : This is fantastic, thanks for sharing.

CCing @sumanjeet0012, @pacrob, @lla-dane, @acul71 and @yashksaini-coder, who are working with me on this front. Wish to have their thoughts and feedback too.

[yashksaini-coder]

I went through the proposal, dug into the codebase, and tried to figure out the best way forward. Here's a rundown.

What's Actually Broken

Right now get_addrs() in BasicHost just asks the transports what addresses they're listening on, sticks /p2p/{peer_id} on each one, and that's what gets handed out to every peer via identify, DHT, rendezvous, etc. If you're behind NAT or on EC2, the transport only sees the private IP. There's no way to tell the host "also advertise this public address." I confirmed this by tracing the code — get_transport_addrs() loops through self._network.listeners and nothing else feeds into it.

Why I Picked Option 1

The proposal had four options. I looked at all of them:

• announce_addrs (option 1) — simplest, covers the main use case (you know your public IP, just pass it in). This is what js-libp2p does.
• address_factory (option 2) — more flexible but more complex. Users have to write a callback. Overkill for now.
• UPnP (option 3) — there's already a UpnpManager in the repo, but UPnP doesn't work on cloud infra and is flaky on home networks.
• ObservedAddrManager (option 4) — the "right" long-term answer but needs a whole confirmation system. Much bigger scope.

Went with option 1. It unblocks the immediate problem and the other options can layer on top later without changing anything.

What I Changed

Four files, ~80 lines of production code, ~60 lines of tests.

• basic_host.py — Added _normalize_announce_addr() to strip /p2p/ suffixes from announce addrs (security: prevents advertising someone else's peer ID). Added announce_addrs keyword-only param to __init__. Modified get_transport_addrs() to merge listen + announce addrs with dedup via a seen set.
• routed_host.py — Same param, just forwards to BasicHost through super().
• __init__.py — new_host() now accepts announce_addrs and passes it to whichever host class it builds.
• test_basic_host.py — Three tests: addrs show up in output, duplicates get removed, /p2p/ suffixes get stripped.

IHost interface is completely untouched. get_addrs() signature unchanged. Everything downstream picks up announce addrs automatically because they flow through get_transport_addrs().

All tests pass (new and existing), default is announce_addrs=None so nothing breaks for anyone who doesn't use it.

Usage

from libp2p import new_host
from multiaddr import Multiaddr

host = new_host(
    listen_addrs=[Multiaddr("/ip4/0.0.0.0/tcp/4001")],
    announce_addrs=[Multiaddr("/ip4/203.0.113.1/tcp/4001")],
)
# get_addrs() now includes both. identify/DHT/rendezvous pick it up automatically.

Things I'd Flag for Later

• _normalize_announce_addr() silently swallows exceptions — should probably log a warning there.
• Tests only go through BasicHost directly. No coverage for the RoutedHost or new_host() paths (they're thin wrappers, but still).
• No negative test for invalid input at construction time.
• Dedup is string-based, which could theoretically miss semantically-equivalent multiaddrs with different serializations. Fine in practice.

Future Options

All three remaining proposal options can build on this without breaking it:

• address_factory → transform step inside get_transport_addrs()
• ObservedAddrManager → merge confirmed addresses into the same list
• UPnP → auto-populate announce_addrs when an external IP is discovered

---

@asabya I would like to take this and work on it.

[asabya]

#1268 I already have a PR for this added.