Re: [Qemu-block] [RFC] QMP design: Fixing query-block and friends

[Alberto Garcia]

On Fri 30 Jun 2017 04:22:11 PM CEST, Kevin Wolf wrote:
> Am 30.06.2017 um 15:01 hat Alberto Garcia geschrieben:
>> On Tue 27 Jun 2017 06:31:45 PM CEST, Kevin Wolf wrote:
>> > * Speaking of recursion: ImageInfo recursively includes information
>> >   about all images in the backing chain. This is what makes the output
>> >   of query-named-block-nodes so redundant. It is also inconsistent
>> >   because the runtime information (BlockInfo/BlockDeviceInfo) isn't
>> >   recursive.
>> 
>> I've also been told of cases where a query-block command on a VM with a
>> very large amount of images in all backing chains would slow down the
>> guest because of this recursion.
>> 
>> I haven't been able to reproduce this problem myself, but being able to
>> see only the devices seen by the guest (as opposed to the complete
>> graph) seems like a good idea.
>
> I think the information only really becomes redundant if you use
> query-named-block-nodes because then you list every backing file node
> and each of them contains the recursive information.
>
> With query-block, you start at the top and include the information for
> each image in the backing file chain only once. If we did indeed have
> a problem there, that might mean that we do in fact need filtering to
> reduce the overhead. But I don't think any of the information involves
> any I/O to get, so it seems unlikely to me that this would make a big
> difference.

Yes, that was also what I thought, but I haven't been able to reproduce
the problem so I don't know yet. In my own tests with thousands of
backing files I haven't noticed any slowdown caused by query-block.

Berto