Re: [Libvirt-cim] Re: KVM on Pegasus Test Run Summary for Sep 10 2008 [ Current Source]
> =================================================
> FAIL Test Summary:
> ComputerSystem - 23_suspend_suspend.py: FAIL
The above tc passed when run manually.
Do you know what caused the guest to fail to start? Does the test
always fail during bulk run?
It sounds like a guest from a previous test isn't being cleaned up
properly. This will need to be fixed.
> =================================================
> XFAIL Test Summary:
> ComputerSystem - 32_start_reboot.py: XFAIL
> ComputerSystem - 33_suspend_reboot.py: XFAIL
> ResourceAllocationFromPool - 05_RAPF_err.py: XFAIL
The above RAFP tc is written to verify that RAFP returns appropriate
error when a non-existing bridge or networkpool is used in defining a
guest.
But, the tc now XFAIL's because it returns a valid RAFP record.
This should be a failure and not an XFAIL. There's no bug number
associated with the failure, so the logic of try_assoc() is incorrectly
returning an XFAIL.
This is due to the use of the cim_define() in the test case.
Before calling the cim_define() we make necessary changes to the XML
configuration to reflect the invalid bridge/networkpool name.
But when we use the cim_define() function we do not have any means where
we can use the invalid networkpool name or bridgename.
cim_define() uses the VSSD and various RASD objects that belong to the
instance of the VirtCIM class that is created. Instead of using
modify_net_name(), the invalid network name should be passed in when the
instance of VirtCIM is initialized.
Hence the DefineSystem() in the cim_define() goes ahead to define a
guest with the valid networkpoolname and RAFP returns a record, which is
against the test case.
I ran the tc for Xen/XenFV and the tc fails there as well.
Can we revert back to the define() [which used virsh to define the
guest], which initially existed in the tc ??
If we use the virsh define, we're testing the virsh define call - the
provider DefineSystem() is not being tested in this case. Wherever
possible, we should be using DefineSystem() so that the provider stack
is tested.
--
Kaitlin Rupert
IBM Linux Technology Center
kaitlin(a)linux.vnet.ibm.com