Hole-punching is not possible during reconnection.

[jclab-joseph]

Under the condition that A [connect to B] ----> [inbound] B peer exists,

Hole punching will not occur if A-peer is restarted.

Because,

• go-libp2p/p2p/protocol/holepunch/holepuncher.go

  Line 279 in 7b78dd6

  func (nn *netNotifiee) Connected(_ network.Network, conn network.Conn) {

  is called, but
• For UDP or unintended disconnection is not detected immediately,

  go-libp2p/p2p/protocol/holepunch/holepuncher.go

  Line 111 in 7b78dd6

  if getDirectConnection(hp.host, rp) != nil {

  incorrectly assumes that a connection already exists.

2025-10-30T23:02:34.648+0900    �[35mDEBUG�[0m  p2p-holepunch   holepunch/holepuncher.go:95     beginDirectConnect      {"host": "12D3KooWLcU8rxuwQphJq7DJGRwPx6qdFY8Aj6thpdjXyTcshEe6", "peer": "12D3KooWBAm5U5jMMeasz15KnP7Tg4fn2Qd68WAc7YzJoSosfPmt"}
2025-10-30T23:02:34.648+0900    �[35mDEBUG�[0m  p2p-holepunch   holepunch/holepuncher.go:112    already connected       {"host": "12D3KooWLcU8rxuwQphJq7DJGRwPx6qdFY8Aj6thpdjXyTcshEe6", "peer": "12D3KooWBAm5U5jMMeasz15KnP7Tg4fn2Qd68WAc7YzJoSosfPmt"}

Version Information

github.com/libp2p/go-buffer-pool v0.1.0
github.com/libp2p/go-cidranger v1.1.0
github.com/libp2p/go-doh-resolver v0.5.0
github.com/libp2p/go-flow-metrics v0.3.0
github.com/libp2p/go-libp2p v0.43.0
github.com/libp2p/go-libp2p-asn-util v0.4.1
github.com/libp2p/go-libp2p-kad-dht v0.34.0
github.com/libp2p/go-libp2p-kbucket v0.7.0
github.com/libp2p/go-libp2p-record v0.3.1
github.com/libp2p/go-libp2p-routing-helpers v0.7.5
github.com/libp2p/go-libp2p-testing v0.12.0
github.com/libp2p/go-libp2p-xor v0.1.0
github.com/libp2p/go-msgio v0.3.0
github.com/libp2p/go-netroute v0.2.2
github.com/libp2p/go-reuseport v0.4.0
github.com/libp2p/go-yamux/v5 v5.0.1
github.com/libp2p/zeroconf/v2 v2.2.0

[MarcoPolo]

I probably don't follow, but if A is restarted all connections to A will be lost, and A will lose its hole-punching state. It would be very hard to support holepunching during a node restart.

[jclab-joseph]

It’s natural to lose the hole-punching state when a node restarts.
However, in such cases, I think a new hole-punching attempt should be made.
This issue occurs because the B-peer doesn’t detect that the existing connection has already been lost.
It would probably be resolved if there were a retry mechanism instead of performing hole-punching only once.