Re: [Qemu-devel] [PATCH v7 2/6] qmp: Create IOThrottle structure

[Eric Blake]

On 08/16/2017 11:13 AM, Markus Armbruster wrote:
> Markus Armbruster <address@hidden> writes:
> 

>>
>> Conclusion: no consensus, yet.
> 
> All right, let's start over and try to resolve the impasse and/or
> misunderstanding.
> 
> Type BlockIOThrottle lives in qapi/block-core.json, and is used by QMP
> command block_set_io_throttle.  Since 1.1.
> 
> Pradeep has a use case for throttling in fsdev.  Instead of duplicating
> the relevant parts of BlockIOThrottle, qmp_block_set_io_throttle() and
> hmp_block_set_io_throttle(), he factors them out smartly, into
> 
> * [PATCH 2] IOThrottle, base type of BlockIOThrottle
> 
> * [PATCH 3] throttle_set_io_limits(), called by
>   qmp_block_set_io_throttle()
> 
> * [PATCH 4] hmp_initialize_io_throttle(), called by
>   hmp_block_set_io_throttle()
> 
> throttle_set_io_limits() goes into existing util/throttle.c, and
> hmp_initialize_io_throttle() goes into existing hmp.c.  The question is
> where IOThrottle should go.

Good summary.

> 
> Pradeep proposes to put it in new qapi/throttle.json.  Certainly
> defensible, but I really don't like putting every little thing shared
> across subsystem boundaries into its own schema file.

I agree with the dislike of creating new files, if an existing file is
adequate.

> 
> Let me step back and discuss why we split the QAPI schema into multiple
> files in the first place.  For me, the one and only reason is
> MAINTAINERS.

Indeed, that's a good description of why splits would be appropriate.
So the obvious next question is if this is a case that needs a new
maintainer.

> 
> If the block folks should continue to maintain IOThrottle, then it
> should stay put in block-core.json.

I think Manos' work on making throttling a filter driver at the block
layer is proof enough that it it is still fine to keep throttling
maintained in block-core.json.

> 
> If somebody else should start maintaining it, it should move.  We'd need
> a suitable entry in MAINTAINERS then.
> 
> I don't see why maintenance should change, and therefore believe it
> should stay put.
> 
> Eric?

I think we're in violent agreement: don't create a new file, and having
the new factored type live in block-core.json is the best fit because we
haven't come up with any reasons why it needs to be split.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

signature.asc
Description: OpenPGP digital signature