Node can end up with an empty discovery table

[oneeman]

Expected Behavior

A node's discovery table should include the bootnode when it starts up, and should not subsequently become empty.

Current Behavior

We have seen cases where a node fails to connect to peers and its discovery table (inspected using admin.discoverTableInfo) was empty. In particular, we saw this on a private testnet when the bootnode passed in using the --bootnode flag had an incorrect enode (did not match the enode actually being used on the bootnode).

A validator also reported being unable to connect to peers on startup, and the logs showed messages such as:

INFO [05-08|17:16:11.344] Looking for peers                        peercount=0 tried=0 static=0

This sounds like it may be due to this same problem, though we do not know for sure whether the discovery table was empty. The validator reports that restarting the node did not fix it, but removing the docker container and creating a new one did

This has happened in the past, but should have been fixed by #1194. However, since this is still being seen sometimes in v1.3.x, it seems there is still a way for this problem to happen.

Steps to Reproduce Behavior

No known repro steps yet.

System information

Seen with v1.3.x