[Piwik-hackers] API: querying data for last X

[Anton Andrievskij]

Hi, Matt. What do you think about my reflections? Any feedback?

Anton

On 6/4/08, Anton Andrievskij <x.meglio at gmail.com> wrote:
>
> Hi, Matt.
>
> I researched on this problem little more and I found real bug in the logic.
> Follow instructions below and you will see where the problem is:
>
> 1. Setup always_archive_data = false
>
> 2. Create new html page, add it as new site to Piwik and put JavaScript
> code to the page. DO NOT query for statistics for this site (actually it is
> only 1 page)
>
> 3. Make some visits from different browsers and/or machines.
>
> 4. Query for last day, then last week, month and for last year. You will
> see the same numbers for visits, actions etc. Do you think it is correct
> now?  You just told me in the previous letter that week, month and year
> statistics do not include today's statistics. But we can see all visits
> here.
>
> Actually Piwik can't find archive built for last week, and it build it
> including today's visits. So Piwik does not follow the same logic. See
> next...
>
> 5. Now you have all archives built. Try to make some more visits and query
> for last day, week, month and year again. You will see new visits in the
> last day, but you will not see them in last month and year, but you will see
> it for last week! So it is totally confusing. Now Piwik does not include new
> visits from today (just like you said in your previous message). But it does
> it if no archives built in the past.
>
> P.S. If you set always_archive_data = true and ask Piwik for visits for
> last day, week, month and year again - it will be ok including all today's
> visits.
>
> So Piwik even does not follow your logic described in the previous letter.
> It seems that the logic is that Piwik checks whether archive is built
> already and simply does not rebuild it. But then it makes problems - you can
> see it in example above.
>
> So what I propose is that you must go over the logic again and fix it. It
> must NEVER include today's statistics... or ALWAYS include today's
> statistics. Or even it is more better to have special configuration like you
> said: include_today_statistics_in_period_archives.
>
> Please make it high priority bug / task because it is error in the Piwik's
> counting logic and I'm compelled to set always_archive_data = true to avoid
> this bug. But it slows Piwik calculation very much. I checked it and I get
> ~4x performance when I set it back to false.
>
> Please we need this update urgently.
>
> Thanks you again,
> Anton Andriyevskyy
>
> On Wed, Jun 4, 2008 at 10:42 AM, Anton Andrievskij <x.meglio at gmail.com>
> wrote:
>
>> Hi, Matt.
>>
>> You wrote:
>> >> It would be incorrect as today's stats are never up to date.
>>
>> Actually when you query visits by weeks, months or years you will always
>> get uncompleted data.
>>
>> Example. If you query for last 3 weeks today, the latest week will be
>> 2008-06-02 to 2008-06-08. What you are doing in Piwik is you do not include
>> today's visits, but in this case the latest week still remains incomplete
>> because there are also future days which are not included in the sum.
>>
>> So when I query for last week - I actually want to know the real value
>> including today's visits - then it makes sense. I think you definitely have
>> to add this option in configuration:
>> include_today_statistics_in_period_archives. Or else querying for last week,
>> month or year have no sense?
>>
>> The real example where it is interpreted as error is when I want to show
>> to users the visits of their pages for last week. They tell me: I asked my
>> friend to open this page and then I'm going to see statistics, but there is
>> still 0 visits. Why? Every people think that it is error and statistics does
>> not work. So I think the idea is to have Piwik working as user expected. It
>> will be correct from user's point of view, not from developer's point of
>> view. What do you think about it?
>>
>> Actually I'm forced to configure Piwik to rebuild archives every time -
>> and then I get correct statistics and testers are happy then. But then I'm
>> worried about speed. Does this configuration really slows Piwik with
>> thousands of users/pages?
>>
>> Thanks for discussion,
>> Anton Andriyevskyy
>>
>> On Wed, Jun 4, 2008 at 4:03 AM, Matthieu Aubry <matthieu.aubry at gmail.com>
>> wrote:
>>
>>> (In the future please post a new message for each distinct question)
>>>
>>> This is the behaviour for piwik archives: period statistics (week / month
>>> / year statistics) do *not* include today's statistics.
>>> It would be incorrect as today's stats are never up to date. This is also
>>> for performance reasons (we have to compute these reports only once a day).
>>>
>>> There could be an option "include_today_statistics_in_period_archives"
>>> but I'm not sure about the value here.
>>>
>>> Anton Andrievskij wrote:
>>>
>>> When I query for last 2 weeks (or last 2 days) - I get today's visits:
>>>
>>> http://mysite.com/piwik/?module=API&method=VisitsSummary.get&idSite=2&period=
>>> *week*&date=last2&format=xml
>>>
>>> But if I get for last 2 months (or last 2 years) - I do not get today's
>>> visits:
>>>
>>> http://mysite.com/piwik/?module=API&method=VisitsSummary.get&idSite=2&period=
>>> *month*&date=last2&format=xml
>>>
>>>
>>> _______________________________________________
>>> Piwik-hackers mailing list
>>> Piwik-hackers at piwik.org
>>> http://lists.piwik.org/cgi-bin/mailman/listinfo/piwik-hackers
>>>
>>>
>>
>>
>> --
>> Best regards,
>> Anton Andriyevskyy
>>
>> Please think about the environment
>> and do not print this e-mail
>> unless you really need to.
>
>
>
>
> --
> Best regards,
> Anton Andriyevskyy
>
> Please think about the environment
> and do not print this e-mail
> unless you really need to.
>



-- 
Best regards,
Anton Andriyevskyy

Please think about the environment
and do not print this e-mail
unless you really need to.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.piwik.org/pipermail/piwik-hackers/attachments/20080606/3cb75422/attachment.htm