Am 28.02.2018 um 20:24 hat John Snow geschrieben:
> 
> 
> On 02/28/2018 01:29 PM, Kevin Wolf wrote:
> > Am 27.02.2018 um 21:24 hat Eric Blake geschrieben:
> >> On 02/23/2018 05:51 PM, John Snow wrote:
> >>> This allows us to easily force the option for all jobs belonging
> >>> to a transaction to ensure consistency with how all those jobs
> >>> will be handled.
> >>>
> >>> This is purely a convenience.
> >>>
> >>> Signed-off-by: John Snow <js...@redhat.com>
> >>> ---
> >>
> >>> +++ b/qapi/transaction.json
> >>> @@ -79,7 +79,8 @@
> >>>   ##
> >>>   { 'struct': 'TransactionProperties',
> >>>     'data': {
> >>> -       '*completion-mode': 'ActionCompletionMode'
> >>> +       '*completion-mode': 'ActionCompletionMode',
> >>> +       '*manual-mgmt': 'bool'
> >>
> >> Missing QAPI documentation (what you have elsewhere in the C code can
> >> probably be copied here, though).
> >>
> >> The UI aspect makes sense (I can declare one manual at the transaction 
> >> level
> >> instead of multiple manual declarations per member level within the
> >> transaction).
> > 
> > I'm not so sure if I like the interface, it duplicates functionality in
> > two places.
> > 
> > At th very least I would make job creation without BLOCK_JOB_MANUAL an
> > error if the transaction requires it instead of silently overriding the
> > option that was given to the individual job. But honestly, it might be
> > better to just leave this one away.
> > 
> > Kevin
> > 
> 
> Sure, I put it in the trailing position here because I see it as
> optional. I don't like the idea of having to specify manual for each and
> every item in a transaction, but if mixed-mode is possible then this is
> less important.

For management tools it shouldn't really matter if they have that one
line of code when creating TransactionProperties or in the loop that
creates the individual jobs.

> I'll leave it off for now, but I will always fondly remember it, and
> then maybe try to sneak it back in for v6.

:-)

If you try to sneak it in, just make sure that conflicting settings
result in an error rather than one of them being silently overridden.

Kevin

Reply via email to