Re: [PATCH v5 11/12] migration/dirtyrate: Implement qmp_cal_dirty_rate()/qmp_get_dirty_rate() function

[Zheng Chuan]

On 2020/8/27 19:58, David Edmondson wrote:
> On Thursday, 2020-08-27 at 17:34:13 +08, Zheng Chuan wrote:
> 
>>>> +    /*
>>>> +     * Only support query once for each calculation,
>>>> +     * reset as DIRTY_RATE_STATUS_UNSTARTED after query
>>>> +     */
>>>> +    (void)dirtyrate_set_state(&CalculatingState, CalculatingState,
>>>> +                              DIRTY_RATE_STATUS_UNSTARTED);
>>>
>>> Is there a reason for this restriction? Removing it would require
>>> clarifying the state model, I suppose.
>>>
>> We only support query once for each calculation.
>> Otherwise, it could always query dirtyrate, but maybe the dirtyrate is 
>> calculated
>> long time ago.
> 
> There's nothing in the current interface that prevents this from being
> the case already - the caller could initiate a 1 second sample, then
> wait 24 hours to query the result.
> 
> Obviously this would generally be regarded as "d'oh - don't do that",
> but the same argument would apply if the caller is allowed to query the
> results multiple times.
> 
> Perhaps a complete solution would be to include information about the
> sample period with the result. The caller could then determine whether
> the sample is of adequate quality (sufficiently recent, taken over a
> sufficiently long time period) for its' intended use.
> 
You mean add timestamp when i calculate?
Actually, I do not want make it complicate for qemu code,
maybe it could be left for user to implement both two qmp commands
like in libvirt-api.

On the other hand, it really bother me that we need to reset calculating state
to make sure the state model could be restart in next calculation.

For now, i put it after query_dirty_rate_info is finished as you see, it should 
not be a good idea:(

Maybe it is better to initialize at the beginning of qmp_calc_dirty_rate().

> dme.
>