[libvirt] [PATCH] qemu: hotplug: unify "not found" logs when detaching device

[John Ferlan]

On 12/15/2017 09:50 PM, Chen Hanxiao wrote:
> 
> 
> At 2017-12-16 09:26:32, "John Ferlan" <jferlan at redhat.com> wrote:
>>
>>
>> On 12/14/2017 06:16 AM, Chen Hanxiao wrote:
>>> From: Chen Hanxiao <chenhanxiao at gmail.com>
>>>
>>> Some services, such as Nova, check whether device was not found
>>> by errror messages "not found". [1]
>>
>> error
>>
>>>
>>> This patch unify logs of qemuDomainDetachDeviceLive, which will be helpful.
>>>
>>> [1] https://github.com/openstack/nova/blob/master/nova/virt/libvirt/guest.py#L406
>>>
>>> Signed-off-by: Chen Hanxiao <chenhanxiao at gmail.com>
>>> ---
>>>  src/qemu/qemu_hotplug.c | 18 +++++++++++-------
>>>  1 file changed, 11 insertions(+), 7 deletions(-)
>>>
>>
>> Something about a tool that parses the error message(s) looking for a
>> specific message string in English and needing to alter libvirt sources
>> to match that tools' needs strikes me as incorrect and a "slippery
>> slope" to follow.
>>
>> I'm not in favor of this because we'll be constantly chasing these types
>> of bugs to match some other tools' (what I think is) incorrect means to
>> handle errors.
>>
> 
> Agree.
> But we don't have enough error code to cover all of scenario.
> 

Then there'd be a "different" fix required - adding an error code...
Perhaps VIR_ERR_NO_DEVICE which would have error text "device %s not found"?

Of course that would require up the stack source code changes as well.
Still better than scanning the error message which only works for a
specific language set. IOW: how would this work for non "English" languages?

> For qemuDomainDetachDeviceDiskLive, VIR_ERR_OPERATION_FAILED can be
> "disk not found", also can be "cannot hot unplug multifunction PCI device"
> in the following call of qemuDomainDetachVirtioDiskDevice.

So it's "OK" to have a different operational failure? I assume that'll
cause a failure elsewhere in the code. I didn't go digging on the source
code - just a quick look at the _try_detach_device method from the above
nova link.

> 
> So the tools powered by libvirt had to find a workaround by 
> analyzing error messages...

Again, wouldn't those tools be broken without the right language set
being used?

John

> 
> Regards,
> - Chen
>