tcp: Listener terminated with error causing litep2p backend to exit

[lexnv]

Some Paseo validators have reported the following error:

Sep 06 15:54:25 vo-phx1 polkadot\[957593\]: 2025-09-06 15:54:25.480 ERROR tokio-runtime-worker litep2p::tcp: TCP listener terminated with error error=Os { code: 103, kind: ConnectionAborted, message: "Software caused connection abort" }
Sep 06 15:54:25 vo-phx1 polkadot\[957593\]: 2025-09-06 15:54:25.480 ERROR tokio-runtime-worker litep2p::transport-manager: Installed transports terminated, ignore if the node is stopping
Sep 06 15:54:25 vo-phx1 polkadot\[957593\]: 2025-09-06 15:54:25.480 ERROR tokio-runtime-worker sub-libp2p: Litep2p backend terminated
Sep 06 15:54:25 vo-phx1 polkadot\[957593\]: 2025-09-06 15:54:25.588  INFO tokio-runtime-worker sub-authority-discovery: Successfully persisted AddrCache on disk
Sep 06 15:54:26 vo-phx1 polkadot\[957593\]: 2025-09-06 15:54:26.370  WARN tokio-runtime-worker parachain::availability-recovery: Recovery of available data failed. candidate\_hash=0xf8be43ec2e00ab7210e6f864e71f5dfcc25b1137b854fc066fcfb7f549aba6ca traceID=>
Sep 06 15:54:26 vo-phx1 polkadot\[957593\]: 2025-09-06 15:54:26.370  WARN tokio-runtime-worker parachain::availability-recovery: Error during recovery: Data is unavailable

The TCP Listener has returned an error, which is propagated to the TCP transport as stream terminated (Some(None)):

litep2p/src/transport/tcp/mod.rs

Lines 572 to 580 in 6723585

	Some(Err(error)) => {
	tracing::error!(
	target: LOG_TARGET,
	?error,
	"TCP listener terminated with error",
	);
	Poll::Ready(None)
	}

The downstream effect of this is that the TCP transport will terminate on a None event. Then, the litep2p backend will terminate on the None event propagated from the TCP transport.

This is effectively causing a cascade of events that terminate the litep2p backend.

The error originates from the Tokio TCP listener:

litep2p/src/transport/common/listener.rs

Lines 460 to 465 in 6723585

	match listener.poll_accept(cx) {
	Poll::Pending => {}
	Poll::Ready(Err(error)) => {
	self.poll_index = (self.poll_index + 1) % len;
	return Poll::Ready(Some(Err(error)));
	}

This might happen under unstable network conditions; however, we should retry the operation or reinitialize the listener.

cc @paritytech/networking

[lexnv]

A similar approach should be ported to the WebSocket implementation

[dmitry-markin]

It might be due to network interfaces being brought down. litep2p should survive such scenario, but it's not something node administrators should do with the interface the node is listening on. If it's due to other reasons we should handle it.

[dmitry-markin]

Was not able to reproduce by bringing down the interface / deleting the address / deleting the interface.