Meta-data service regularly stalls/errors

[mortrom]

This may or may not be related to my other post (though don’t normally happen together);

Azure hosts disconnecting regularly

I’m experiencing Rancher hosts disconnecting every few days, and with about 20 of them across 4 environments that can mean I can expect one or more almost every day. Restarting the agent always rectifies. Here’s some errors that are occurring in the agent logs prior to the disconnect; time="2018-03-19T15:27:44Z" level=error msg="Received error reading from socket. Exiting." error="read tcp 10.33.46.4:57782->10.33.250.119:80: read: connection reset by peer" time="2018-03-19T15:27:4…

Quite regularly I’m experiencing the meta-data service either stalling (or at least not outputting any logs) or unable to discover self with errors around such. Sadly I don’t have an example of the latter right now. I’ll add if when it happens again.

Restarting the network services stack will generally rectify the situation. Restarting the one host specific meta-data service in question sometimes works, but stopping and starting the stack is more reliable.

Any ideas?

Here’s the Environment that is running;

Rancher - 1.6.12
Rancher Agent - 1.2.7
Rancher Container Orchestration - Cattle
Docker - 17.06.2-ce
Host/Server OS - Ubuntu 16.04.3 LTS
Cloud - Azure