Open Stack

On Sun, Mar 22, 2015 at 04:58:00PM +0000, Moshe Levi wrote:
> Hi,
> 
> In the latest master nova code I am keep getting this error  "An object of type PciDevicePoolList is required here"
> 
> My nova.conf contains  pci_passthrough_whitelist.
> 
> When I tried to launch vm after devstack installation the vm was successfully booted.
> When I restart the compute node and then try to launch vm I get a failure due to  error "An object of type PciDevicePoolList is required here". (It doesn't matter if it vm with normal or vm with direct port )
> 
> In the debugger  I can see the in that one of resources sent to the scheduler is "pci_device_pools" which is a list for example ('pci_device_pools': [{'count': 7, 'vendor_id': u'15b3', 'product_id': u'1004', 'tags': {u'numa_node': None, u'physical_network': u'physnet1'}}])
> When this resource saved into the database I get  the above error.
> Please note I can reproduce this issue only after I restart the compute node.
> Removing the pci_device_pools key from the resources (remove it from self.compute_node in the resource_tracker) fix this issue, but I am not sure that it is the correct way to go.
> 
> Is anyone see  this issue?
> Should the pci_device_pools be sent to the scheduler?

It's actually not being sent to the scheduler (even though, confusingly,
the call is to nova.scheduler.client.report.update_resource_stats()).
It's actually just going to the conductor, and then the database.

Looks like, indeed, there was a bug introduced recently. Have you filed
a bug yet? If not, please do and we'll get to work on it.

Best,
-jay