Stop trying to connect to yourself and spamming the logs

[tomwilkie]

I start weave and pass every peer the same set of IPs, which happens to include their own. The logs are full of messages like this:

INFO: 2015/08/16 10:28:26.594151 ->[172.31.13.96:6783] attempting connection
INFO: 2015/08/16 10:28:26.594664 ->[172.17.42.1:50625] connection accepted
ERRO: 2015/08/16 10:28:26.595323 ->[172.31.13.96:6783|fe:09:5f:88:03:b4(ip-172-31-13-96)]: connection shutting down due to error: Cannot connect to ourself
ERRO: 2015/08/16 10:28:26.595593 ->[172.17.42.1:50625|fe:09:5f:88:03:b4(ip-172-31-13-96)]: connection shutting down due to error: Cannot connect to ourself
INFO: 2015/08/16 10:33:44.324598 ->[172.31.13.96:6783] attempting connection
INFO: 2015/08/16 10:33:44.324989 ->[172.17.42.1:50638] connection accepted
ERRO: 2015/08/16 10:33:44.325565 ->[172.31.13.96:6783|fe:09:5f:88:03:b4(ip-172-31-13-96)]: connection shutting down due to error: Cannot connect to ourself
ERRO: 2015/08/16 10:33:44.325803 ->[172.17.42.1:50638|fe:09:5f:88:03:b4(ip-172-31-13-96)]: connection shutting down due to error: Cannot connect to ourself

Could we either remember that the IP pointed to ourself, and stop trying to connect to it, or only log errors if they are different from last time?

[bboreham]

As noted at #611, the weave router is running inside a container and containers do not generally know their externally-visible address. And NAT mappings can change so it is dangerous to cache the results of a particular connection. Although perhaps a mapping to ourself is unlikely to change.

Only logging errors when they change is an interesting idea.

[rade]

perhaps a mapping to ourself is unlikely to change.

That.

Only logging errors when they change is an interesting idea.

We need the error log message in order to 'close' the connection lifecycle in the logs. Also, surely the 'attempting/accepted' log messages are just as annoying. But suppressing those would mean not logging anything for some connections, which is a bad idea IMO.