fleet has no IP information when automatically started on EC2

[bcwaldon]

I booted a five-node CoreOS cluster on EC2 and brought fleet up automatically. Looking at the machine list, I see no IP information:

% fleetctl list-machines
MACHINE     IP      METADATA
18805c5d... -       -
28d9993e... -       -
3a74f97d... -       -
758fe95c... -       -
ade94405... -       -

Restarting fleet on each of the instances does correct the problem:

% fleetctl list-machines
MACHINE     IP      METADATA
18805c5d... 10.80.83.51 -
28d9993e... 10.114.167.102  -
3a74f97d... 10.77.63.132    -
758fe95c... 10.194.243.63   -
ade94405... 10.203.65.129   -

It seems like we're starting fleet before the instances actually have IPs bound to their interfaces, and fleet won't refresh it's machine state once it starts. The Machine object will simply get the IP address at creation time: https://github.com/coreos/fleet/blob/master/machine/machine.go#L24

[polvi]

One solution would be to do some sort of After= on the unit with regards to networking. Not sure how to do this, but should be possible.

[philips]

Lets start by adding a timer to poll the network information. We can make it evented later.

[polvi]

@philips seems reasonable. I think we'll also have to add a conf option to say what interface to use as well. Right now I think it just picks the first from the list.

[bcwaldon]

We already have a timer driving the heartbeat every 15s - how about we just use that?

[philips]

sure, yea, that seems fine.