[rhos-list] EXTERNAL: Re: Setting up Compute Nodes.

[Minton, Rich]

I found a strange occurrence when trying to bring up an instance on the second compute node. While watching the progress on the dashboard, I could see the instance try to come up on the second compute node then it would jump back to the first. I checked the logs and I was getting an "access denied" error when trying to access the instance directory. I did some snooping and found that the directory for my instances was not being mounted. I have my /var/lib/glance/images and /var/lib/nova/instances mounted to an NFS share on my Isilon storage cluster.  The mounts on my controller node work fine and the config files for the automounter are identical on each node.  I was finally able to get it to work by stopping the autofs service and running "automount" in the foreground and in debug mode (just to watch what was going on). Then all my mounts work fine and I can get an instance to come up on the second compute node. Not sure what's going on with the autofs service. I ran yum update on both nodes to make sure everything was in sync version wise.

Has anyone seen this before?

Thank you!
Rick

Richard Minton
LMICC Systems Administrator
4000 Geerdes Blvd, 13D31
King of Prussia, PA 19406
Phone: 610-354-5482



-----Original Message-----
From: Russell Bryant [mailto:rbryant at redhat.com] 
Sent: Wednesday, February 27, 2013 4:33 PM
To: Minton, Rich
Cc: rhos-list at redhat.com; Andrus, Gregory
Subject: Re: EXTERNAL: Re: [rhos-list] Setting up Compute Nodes.

On 02/27/2013 04:20 PM, Minton, Rich wrote:
> Yes, I try to force it to run on compute1 but it runs on controller.
> 
> From the command line I run...
> 
> nova boot --image <image-id> --flavor 3 --key_name test 
> --availability-zone=nova:compute1 server-name

Thanks for the command.

This method of forcing a host is only supported if you're an admin.  It would be ignored for a regular user.  Since the instance did actually start, I assume it was a regular user.  In that case, the 'compute1'
part would be ignored and just the availability zone of 'nova' would be honored.  Since both nodes are in the 'nova' zone, it happened to pick your controller.

There aren't any ways to force an instance to a specific host that don't require some level of admin interaction that I know of.  This really comes down to a core philosophy of this type of system, which is that the user of the cloud shouldn't have to know or care which specific node an instance runs on.  This is not to be confused with nodes having capabilities and requesting instances that have requirements for certain capabilities.  We have various features in that area.

--
Russell Bryant