Re: [Qemu-devel] [PATCH 0/3] Export machine type deprecation info through QMP

[Eduardo Habkost]

On Fri, May 10, 2019 at 08:28:04AM +0200, Markus Armbruster wrote:
> Eduardo Habkost <address@hidden> writes:
> 
> > On Thu, May 09, 2019 at 05:08:11PM +0100, Daniel P. Berrangé wrote:
> >> On Thu, May 09, 2019 at 12:52:47PM -0300, Eduardo Habkost wrote:
> >> > On Thu, May 09, 2019 at 10:14:52AM +0100, Daniel P. Berrangé wrote:
> >> > > On Thu, May 09, 2019 at 10:31:46AM +0200, Markus Armbruster wrote:
> >> > > > We've wandered into the QAPI vs. QOM swamp.  Cc: Paolo.
> >> > > > 
> >> > > > Eduardo Habkost <address@hidden> writes:
> >> > > > 
> >> > > > > On Wed, May 08, 2019 at 11:16:50AM +0200, Markus Armbruster wrote:
> [...]
> >> > > > >> I agree we should point to a preferred replacement whenever we 
> >> > > > >> deprecate
> >> > > > >> something.
> >> > > > >> 
> >> > > > >> We have to do it in documentation.  And we generally do, in
> >> > > > >> qemu-deprecated.texi.
> >> > > > >> 
> >> > > > >> How useful would doing it in QMP as well be?  Depends on what 
> >> > > > >> management
> >> > > > >> applications can do with the additional information.
> >> > > > >
> >> > > > > I expect it to be useful for things that have obvious
> >> > > > > replacements, like old machine type or CPU model versions.
> >> > > > 
> >> > > > I doubt a management application should apply suggested replacements
> >> > > > automatically, and I doubt libvirt would.  Not even when QEMU 
> >> > > > developers
> >> > > > deem them "obvious".
> >> > > 
> >> > > We certainly won't apply the suggested replacement as in many cases
> >> > > it is not going to be a functionally equivalent drop-in.
> >> > 
> >> > Who's "we"?
> >> 
> >> I was refering to libvirt here.
> >> 
> >> > > If QEMU logs it to stderr, it will end up in the per-VM log file
> >> > > libvirt has under /var/log/libvirt/qemu/$GUESTNAME.log.  If QEMU
> >> > > doesn't log it to stderr, then libvirt would just write it to
> >> > > that same log file itself.
> >> > > 
> >> > > If libvirt gains some API or event for notifying apps of deprecation
> >> > > we might bubble it up to the mgmt app that way.
> >> > > 
> >> > > I still feel it is useful to have the suggested replacement in the
> >> > > logs, rather  than only leaving it in qemu-deprecated.texi.  This
> >> > > way the info is immediately visible to both app developers and any
> >> > > support person dealing with bugs.
> >> > > 
> >> > > If the app dev see the suggested replacement upfront they're more
> >> > > likely to make an immediate decision to update their code if the
> >> > > suggestion is trivial. If they need to go find the QEMU docs to
> >> > > lookup what action is required I feel they'll more likely just
> >> > > put the item on their long todo list where it will languish.
> >> > 
> >> > Agreed.  However, note that the audience for deprecation
> >> > information is not just developers and support people.  End users
> >> > need to know when they are relying on a deprecated feature, and
> >> > applications should make it as easy as possible for them to
> >> > update their configurations.
> >> > 
> >> > I'm not suggesting the alternative would be applied
> >> > automatically.  But having the alternative available in a
> >> > machine-friendly way may be the difference between a unhelpful UI
> >> > that just tells the user there's some problem but can't give a
> >> > solution, and one that can really assist the user to fix the
> >> > problem.
> >> 
> >> For some aspects of QEMU it might be possible, but considering the
> >> broader set of things which can be deprecated, I don't think it is
> >> possible to expose a machine consumable "suggestion".
> >> 
> >> Consider the deprecation of the ACL feature. We deprecated monitor
> >> commands "acl_add", "acl_policy", "acl_reset", etc. The suggested
> >> replacement is to use one of the many possible QAuthZ types combined
> >> with the -object arg. Even if we invented some way to express this
> >> in the schema, I don't think any app would usefully consume it.
> >
> > No problem, we don't need to suggest a machine consumable
> > alternative for everything.
> 
> Sure, but we need to get enough value out of it to justify its cost.
> 
> > I'm thinking about features that are visible to the end user and
> > require user action to fix their configuration, like machine type
> > versions or CPU model versions.
> 
> Even those may need translation as we cross layers of the stack.
> 
> The fewer cases we have where the machinery for machine-readable
> deprecation advice is actually useful, the worse its cost / benefit
> ratio is going to be.

Are you arguing the cost is unreasonably large?  I still don't
see what the problem is, here.

-- 
Eduardo