fast catchup (goal node catchup) hangs when running with `EnableP2P: true`

[algorandskiy]

goal node  -d mainnet-p2p catchup
Fast catchup to 47620000#AJQDWWQWWJ6D6Q74AIH26HCYZLZRT2AV2K6RCI54DQI43R3OPIWQ is about to start.
Using external catchpoints is not a secure practice and should not be done for consensus participating nodes.
Type 'yes' to accept the risk and continue: yes

Config:

{
        "EnableP2P": true,
        "BaseLoggerDebugLevel": 5,
        "EnableDHTProviders": true
}

Observed behavior:

1. Node is non-responsible - goal node -d mainnet-p2p status not working either with pending goal node catchup in a console, or after quitting it
2. node.log indicates it is catching up

[algorandskiy]

Intermediate analysis:

1. algod collects available p2p nodes with archival=true from DHT since there is no DNS published p2p archival nodes
2. some of these nodes allow tcp/4190 connection but fail to respond to identity traffic. Then libp2p code continues its new stream logic and hangs indefinitely.

Here is is a log excerpt using one of such nodes (this is non-archival but illustrates the issue well):

export GOLOG_LOG_LEVEL=debug
export GOLOG_OUTPUT=stdout
go run ./cmd/catchpointdump net -n mainnet.algorand.network -p /dns4/r-pn.algorand-mainnet.network/tcp/4190/p2p/12D3KooWAaPiWHx1CZcpZ4rQSDcpzsfMN5ecReLpWh5ZuReoxcnM -r 47750000

downloading catchpoint from /dns4/r-pn.algorand-mainnet.network/tcp/4190/p2p/12D3KooWAaPiWHx1CZcpZ4rQSDcpzsfMN5ecReLpWh5ZuReoxcnM /v1/mainnet-v1.0/ledger/sfg4w
2025-03-05T10:40:33.925-0500    DEBUG    basichost       basic/basic_host.go:805 host 12D3KooWNVdvivtrsbdyuLJjM1gvH2z8zSD7F4egbiGszz2nMwjZ dialing 12D3KooWAaPiWHx1CZcpZ4rQSDcpzsfMN5ecReLpWh5ZuReoxcnM
2025-03-05T10:40:33.971-0500    DEBUG    swarm2  swarm/swarm_dial.go:593 12D3KooWNVdvivtrsbdyuLJjM1gvH2z8zSD7F4egbiGszz2nMwjZ swarm dialing 12D3KooWAaPiWHx1CZcpZ4rQSDcpzsfMN5ecReLpWh5ZuReoxcnM /ip4/167.235.65.180/tcp/4190
2025-03-05T10:41:04.459-0500    INFO     net/identify    identify/id.go:457      failed negotiate identify protocol with peer    {"peer": "12D3KooWAaPiWHx1CZcpZ4rQSDcpzsfMN5ecReLpWh5ZuReoxcnM", "error": "i/o deadline reached"}
2025-03-05T10:41:04.459-0500    WARN     net/identify    identify/id.go:431      failed to identify 12D3KooWAaPiWHx1CZcpZ4rQSDcpzsfMN5ecReLpWh5ZuReoxcnM: i/o deadline reached
2025-03-05T10:41:04.460-0500    DEBUG    basichost       basic/basic_host.go:822 host 12D3KooWNVdvivtrsbdyuLJjM1gvH2z8zSD7F4egbiGszz2nMwjZ finished dialing 12D3KooWAaPiWHx1CZcpZ4rQSDcpzsfMN5ecReLpWh5ZuReoxcnM
2025-03-05T10:41:04.460-0500    DEBUG    swarm2  swarm/swarm.go:476      [12D3KooWNVdvivtrsbdyuLJjM1gvH2z8zSD7F4egbiGszz2nMwjZ] opening stream to peer [12D3KooWAaPiWHx1CZcpZ4rQSDcpzsfMN5ecReLpWh5ZuReoxcnM]

[algorandskiy]

Client side: libp2phttp's RoundTripper uses http.Request's context when establishing a new stream. Catchpoint code sets it to a huge value enough for downloading a few GB snapshot over relatively slow connection. Fixed by forking RoundTripper implementation with reasonable timeout.

Server side: looks like these nodes were configured with low incoming connection limit that was reached.