libp2p not shutting down properly

[jacobheun]

• Version: 0.29.0
• Platform: All
• Subsystem: Shutdown, dialer, latency monitor

Type:

Severity: Med

Description:

When libp2p is stopped the dial queue is not being properly emptied. This can cause in progress outbound dials to not be properly terminated which will hold the process open.

Additionally, the latency monitor is not being stopped by the connection manager which will cause timers to continue.

Steps to reproduce the error:

See ipfs/js-ipfs#3191 (comment)

[vasco-santos]

This seems a side effect of the /dnsaddr resolution per libp2p/js-libp2p-tcp#142

The difference of using /dnsaddr seems to be that it will result in a set of other addresses to dial and the dial queue will be working when the stop happens.

What is happening is:

• libp2p stop calls Dialer destroy: https://github.com/libp2p/js-libp2p/blob/master/src/dialer/index.js#L88
• Dialer destroy aborts all in progress dials
• But a /dnsaddr was still being resolved and blocked on https://github.com/libp2p/js-libp2p/blob/master/src/dialer/index.js#L110
• Once the /dnsaddr is resolved, a new dial target is created and a dial happens, which makes the process on going

So, what we basically need is to make the resolve abortable, keep a record of the on going resolves and abort all of them on Dialer.destroy