GCE bootstraps but fails to provision

[John A Meinel]

I just set up a 'free trial' with GCE, and I was able to enable the Google Compute Engine.

After doing so, I did "juju add-credential google" and passed in the JSON file credentials. I was then able to do "juju bootstrap ... google" just fine. The machine came up as expected.

However, I then did "juju switch controller; juju enable-ha" because I wanted to test it. I can see that there are 2 machines up and running in the GCE dashboard, but there are not *3* machines. Further if I do: "juju status" it reports that 1 machine is running, but 2 machines are still "pending".

There is no /var/lib/juju on the second machine, though cloud-init-output.log looks like it was trying to do something.

I *can* SSH to the second machine that was started, though there is no 3rd machine, and I don't see a jujud running.

Mark also noted the same problem (bootstrap works, provisioning new machines does not).

[John A Meinel (jameinel)]

I checked, and I can manually start 3 instances via the web API. I don't know if they are a different CPU level, or some other difference, but it is possible to create more than 2 instances.

[Curtis Hovey (sinzui)]

GCE free accounts do have enough resources to really use juju. Only by specifying constraints to limit jujus you can you deploy a trivial stack. It should be nice if juju could tell you you don't have enough resources to complete your action instead of trying and failing.

[Cheryl Jennings (cherylj)]

I was able to enable-ha with the current tip of master with a GCE free account. Trying beta7 to see if I can reproduce the failure.

[Cheryl Jennings (cherylj)]

Yeah, this was fixed in beta9. It appears that the OpenPort call was hanging when trying to provision the additional controllers. We now only open the port for the first controller.