Re: [RFC 4/6] migration: Deprecate -incoming <uri>

[Peter Xu]

On Tue, Jun 20, 2023 at 01:10:55PM +0100, Daniel P. Berrangé wrote:
> In some cases it is worth having a convenience option for user friendliness.
> 
> In this case, however, users are already needing to use QMP/HMP on the
> source side to set migration parameters. I think it is reasonable to say
> that doing *exactly* the same with QMP/HMP on the destination side is
> actually more friendly than requiring use of -global on the dest side
> which is different syntax.

The -global parameters also work for the src side for my scripts, so I only
need to attach "-incoming tcp:XXX" for dst cmdline here.

> 
> We don't document the way to use -global with migration parameters so
> when people see '-incoming' I think we are steering them into a trap,
> making it look like -incoming is sufficient on its own. Hence that user's
> mistake recently about not setting migration parameters.
> 
> Overall I agree with deprecating  '-incoming' != 'defer', as I think it
> will better guide users to following the correct steps, as is not a
> usability burden as they're already using QMP/HMP on the source side.

I'd say -global is definitely not for users but for developers.  Just like
we keep around HMP - I'm not sure whether it's used in productions, I
thought it was only for developers and we don't deprecate it eagerly.

No strong opinions here if everyone wants to get rid of that, but before
that I hope we can have some kind of cmdline tool that can easily tunnel
qmp commands so whoever developers using pure cmdlines can switch to the
new way easily.

Thanks,

-- 
Peter Xu