← Back to team overview

openstack team mailing list archive

Re: New instances stuck in BUILD

 

Dear all,

I managed to solve this.

If you are using qpid, you need

rpc_backend=nova.rpc.impl_qpid
qpid_hostname=<ip>

if you are using rabbit, you need
rabbit_host=<ip>

to ensure that the correct amqp is set up.

Cheers
- WP


On Thu, Apr 25, 2013 at 7:44 PM, YIP Wai Peng <yipwp@xxxxxxxxxxxxxxx> wrote:

> Dear all,
>
> I am running openstack grizzly on 2 nodes, with multi-host options and
> without conductor.
>
> Node 1 has compute, api, network, etc
>
> Node 2 has compute, network
>
> When launching new instances, those that are launched on the 1st node
> launches fine. Those not on node 2 gets stuck in BUILD. "nova show xxx"
> prints the following (snipped)
>
> | status                              | BUILD
>              |
> | updated                             | 2013-04-25T11:37:29Z
>             |
> | OS-EXT-STS:task_state               | scheduling
>
> I've been reviewing the logs and there are no errors. I have a few
> questions and hope someone can help.
>
> 1) How do I start tracing this error?
>
> 2) In multi-host mode, are the compute nodes supposed to connect to AMQP
> of the main node, or themselves? I specified rabbit_host and it still shows
> nova-compute connecting to localhost.
>
> 3) In scheduler.log, there is the following
> 2013-04-25 19:37:29.064 WARNING nova.scheduler.host_manager
> [req-2272a45e-8e8a-4af0-96cd-b8a387e051fa 91cd099661f54df7bad0c7248c0451ca
> c91d0ed52321439c8b7d0c05b89bb0c8] No service for compute ID 2
> Is this what is causing the problem?
>
> 4) If I were to use conductor in multi-host, do I need a conductor for
> each compute node? If not, how do I specify where is the conductor?
>
> Any advice is greatly appreciated.
>
> Thanks,
> WP
>

References