Bug Report: Stuck Goroutine in libp2p-pubsub handleSendingMessages

[frystal]

Description:

A critical issue has been identified in the libp2p-pubsub library. Specifically, in the process of a peer publishing a new RPC message to other peers through the function "handleSendingMessages", the function attempts to send messages on an existing stream towards remote peers.

However, if the libp2p-pubsub is configured to use the default "Yamux" as the muxer, the "Write" function operates synchronously without a time limitation. In the event that the write operation in "go-yamux" becomes stuck, it leads to the entire goroutine of "handleSendingMessages" becoming stuck. Consequently, any remaining messages destined for the target remote peer will be dropped due to the queue becoming full.

Expected Outcome:

The libp2p-pubsub library should handle situations where the Write operation becomes stuck in "go-yamux" without causing the entire goroutine of "handleSendingMessages" to become stuck.

Actual Outcome:

The current implementation results in the entire goroutine of "handleSendingMessages" getting stuck, leading to dropped messages due to a full queue.

Environment:

[email protected]

[vyzo]

Sure, but if the sender is stuck you won't be able to send messages anyway.

What do you propose to do?

[frystal]

Could you please assign it the 'bug' label?

[vyzo]

Done, but I am not sure we can resolve this. I will welcome a pr.