quantum 420 error leaves instance in BUILD state

[Aaron Lee]

We setup nova to use the QunatumManager, calling out to quantum for network setup. Quantum was not configured properly, and returned a 420 "Network None could not be found". This bubbled up and left the instance in BUILD state. The solution to this could be to force Quantum to use proper HTTP codes, if that triggers the proper logic in Nova.

[dan wendlandt (danwent)]

Hi Aaron, was this issue handled by the patch you submitted a day after filing this (https://review.openstack.org/#change,1315), or is there still outstanding work to be done?

If so, we can definitely get that work scheduled for essex-2.

[Aaron Lee (aaron-lee)]

That patch was accepted two days ago, and should catch the general case of an exception within instance creation. This does not catch the specific problem of quantum returning non-HTTP error codes.

[dan wendlandt (danwent)]

Ok, let's make sure that QuantumManager properly handles different types of errors that may be returned by Quantum, and that such errors result in reasonable error messages and leave the VM in a the correct state. I believe our current plan is to consider the build of a VM failed if we cannot attach it to the network.

My understanding is that it is perfectly valid for a service like Quantum to define its own error codes within the 4XX range, so I don't think that is the fundamental problem. Its really just that I should have done a better job of error handling :)

[Thierry Carrez (ttx)]

I think that doesn't affect nova code itself, but rather code shipped (atm) with Quantum. Please reopen if you disagree.

[dan wendlandt (danwent)]

Actually, the corresponding change will likely be in the QuantumManager, which is a Nova Network Manager and part of the nova code base, so we should probably keep tracking this in Nova.

[dan wendlandt (danwent)]

Brad, should this be updated to "in progress"? If you don't have a patch already, we should probably move this to E-3. Thanks.

[Brad Hall (bgh)]

We should probably move to e3.. I don't have a patch yet, unfortunately.

[dan wendlandt (danwent)]

Brad, do you have a patch for this, or should I shop around and see if someone else can take it? I know you have a bunch on your plate already for E-3.

[Brad Hall (bgh)]

We should give it to someone else if possible -- its going to be a busy week

[Aaron Lee (aaron-lee)]

Brad, I'll look into this tomorrow morning.

[Brad Hall (bgh)]

Great, thanks Aaron

[Aaron Lee (aaron-lee)]

It looks like this will get fixed with Salvatore's patch; https://review.openstack.org/#change,3101

[dan wendlandt (danwent)]

yes, but salv's patch is only for API v1.1, and the nova code uses v1.0.
 (v1.1 is not finished yet). I'm hoping to cut over to v1.1 in nova during
E-4, but we'll see what kind of changes we can get in during E-4, since
there is technically a freeze.

dan

On Wed, Jan 18, 2012 at 7:40 AM, Aaron Lee <email address hidden>wrote:

> It looks like this will get fixed with Salvatore's patch;
> https://review.openstack.org/#change,3101
>
> --
> You received this bug notification because you are a member of Netstack
> Core Developers, which is subscribed to quantum.
> https://bugs.launchpad.net/bugs/885265
>
> Title:
> quantum 420 error leaves instance in BUILD state
>
> Status in OpenStack Compute (Nova):
> New
> Status in OpenStack Quantum (virtual network service):
> New
>
> Bug description:
> We setup nova to use the QunatumManager, calling out to quantum for
> network setup. Quantum was not configured properly, and returned a 420
> "Network None could not be found". This bubbled up and left the
> instance in BUILD state. The solution to this could be to force
> Quantum to use proper HTTP codes, if that triggers the proper logic in
> Nova.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/nova/+bug/885265/+subscriptions
>

--
~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dan Wendlandt
Nicira Networks: www.nicira.com
twitter: danwendlandt
~~~~~~~~~~~~~~~~~~~~~~~~~~~

[dan wendlandt (danwent)]

I don't think brad is going to get to this one, so probably best to move it out of E-3, unless someone else is planning on doing it.

[Brian Waldon (bcwaldon)]

Backed status down to Confirmed as this can't be In Progress without an asignee

[dan wendlandt (danwent)]

Aaron, is it correct that this is no longer an issue now that Nova uses API 1.1 to contact quantum, meaning that we should be returning standard "not found" codes?

[Aaron Lee (aaron-lee)]

Yes, this is no longer an issue.