Dan,
The issue is that MediaViewer always sends two images, the one a user clicked on and the next in that same article. The second image might be shown to the user later when she clicks the right arrow, at which point a new image is prefetched. So only later it will become clear if the image is actually shown to the user.
We thought of differentiating between explicitly asked first image and implicitly sent follow-up images by adding a new x-analytics parm for prefetched images (and just plainly ignore those), but that would harm our server cache, as two versions of same image would be stored due to slightly different urls.
Several variations are still under discussion, different moments to send a beacon from the client, or add a hook in php.
One thing I haven't proposed yet is to patch the server cache code so that it ignores that extra argument when it decides if the cache needs updating, but still logs the original url. But since I'm not familiar with that environment and we have so many ideas still under review, I'll just drop it here. J And who knows what side effects that would have.
Erik
From: analytics-bounces@lists.wikimedia.org [mailto:analytics-bounces@lists.wikimedia.org] On Behalf Of Dan Andreescu
Sent: Thursday, February 05, 2015 21:46
To: A mailing list for the Analytics Team at WMF and everybody who has an interest in Wikipedia and analytics.
Subject: Re: [Analytics] Virtual file view hack for Media Viewer views
Is this a potential solution to Oliver's concern:
For "real" image views, add an X-Analytics header value of "real-view=true" to the request itself?
If that's not feasible, we should look into using statsv for this (not sure how that works) or having this be a different kafka topic and not consumed into HDFS.
On Thu, Feb 5, 2015 at 11:59 AM, Toby Negrin <tnegrin@wikimedia.org> wrote:
On Thu, Feb 5, 2015 at 8:51 AM, Toby Negrin <tnegrin@wikimedia.org> wrote:
It turns out that the media viewer (on desktop; don't know about mobile) does a lot of caching so just because an image is loaded from swift, it doesn't mean it is viewed. We'd like to provide more accurate stats to the GLAM folks, so yes, I think this needs to be added eventually. Let's leave it out of scope for now.
-Toby
On Thu, Feb 5, 2015 at 8:46 AM, Oliver Keyes <okeyes@wikimedia.org> wrote:
We want to include these files in the pageview definition? :/.
My point was more that we should try to avoid traffic-generating
requests that exist solely as a hack for analytics purposes; it's
artificial work for both users and us. If this is the only way of
doing things that's totally fine.
On 5 February 2015 at 11:38, Toby Negrin <tnegrin@wikimedia.org> wrote:
> Hi Gergo -- I like this idea. As far as capacity, any EL-Hadoop based
> solution would be basically doing the same thing as you propose.
>
> Can you please run it past ops (especially the 404 v 204) part?
>
> Oliver -- the issue is that we'd like to figure out a way to provide
> accurate views of the media files; because of client side caching, we can't
> use the current requests. But your point is a good one -- we'll need to add
> this to the PV definition.
>
> -Toby
>
> On Thu, Feb 5, 2015 at 5:18 AM, Oliver Keyes <okeyes@wikimedia.org> wrote:
>>
>> A nice theory, but if they appear in the webrequest table (presumably
>> they would, and we're not creating an entirely new set of varnishes
>> for the transmission of dummy images?) they have to be factored in.
>> Again, however, the new definition automatically filters them by
>> checking the webrequest source and MIME type, so this is not a
>> problem, as I originally stated.
>>
>> On 5 February 2015 at 08:10, Erik Zachte <ezachte@wikimedia.org> wrote:
>> > Oliver, this is not about pageviews, but about media file views.
>> >
>> >
>> >
>> > These will be collected and dumped separately, as per
>> >
>> > https://www.mediawiki.org/wiki/Requests_for_comment/Media_file_request_counts
>> > .
>> >
>> >
>> >
>> > Erik
>> >
>> >
>> >
>> >
>> >
>> > From: analytics-bounces@lists.wikimedia.org
>> > [mailto:analytics-bounces@lists.wikimedia.org] On Behalf Of Nuria Ruiz
>> > Sent: Wednesday, February 04, 2015 22:28
>> > To: A mailing list for the Analytics Team at WMF and everybody who has
>> > an
>> > interest in Wikipedia and analytics.
>> > Subject: Re: [Analytics] Virtual file view hack for Media Viewer views
>> >
>> >
>> >
>> >>We would add a rule to Vagrant to make sure it does not try to look up
>> >> such
>> >> requests in Swift but returns a 404 immediately.
>> >
>> > I bet ops would like it a lot better if this is a 204 and it kind of
>> > makes
>> > sense as it is the code used for beacons and such. Otherwise they might
>> > get
>> > alarms on 404s increasing.
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> > On Wed, Feb 4, 2015 at 12:38 PM, Oliver Keyes <okeyes@wikimedia.org>
>> > wrote:
>> >
>> > Not really; the new pageviews definition wouldn't include those files
>> > anyway. It seems silly, thought, be deliberately generating a large
>> > amount of automated noise and client requests for this :/.
>> >
>> >
>> > On 4 February 2015 at 15:00, Gergo Tisza <gtisza@wikimedia.org> wrote:
>> >> Hi all,
>> >>
>> >> Erik Zachte is working on file view stats and is looking for a way to
>> >> track
>> >> Media Viewer image views (for which there is no 1:1 relation between
>> >> server
>> >> hits and actual image views); after some back and forth in
>> >> https://phabricator.wikimedia.org/T86914 I proposed the following hack:
>> >>
>> >> whenever the javascript code in MediaViewer determines that an image
>> >> view
>> >> happened (e.g. an image has been displayed for a certain amount of
>> >> time),
>> >> it
>> >> makes a request to a certain fake image, say
>> >>
>> >> upload.wikimedia.org/wikipedia/commons/thumb/0/00/Virtual-imageview-<real
>> >> image name>/<size>px-thumbnail.<ext> . These hits can than be easily
>> >> filtered from the varnish request logs and added to the normal
>> >> requests.
>> >> We
>> >> would add a rule to Vagrant to make sure it does not try to look up
>> >> such
>> >> requests in Swift but returns a 404 immediately.
>> >>
>> >> This would be a temporary workaround until there is a proper way to log
>> >> virtual image views, such as EventLogging with a non-SQL backend.
>> >>
>> >> Do you see any fundamental problem with this?
>> >>
>> >
>> >> _______________________________________________
>> >> Analytics mailing list
>> >> Analytics@lists.wikimedia.org
>> >> https://lists.wikimedia.org/mailman/listinfo/analytics
>> >>
>> >
>> >
>> >
>> > --
>> > Oliver Keyes
>> > Research Analyst
>> > Wikimedia Foundation
>> >
>> > _______________________________________________
>> > Analytics mailing list
>> > Analytics@lists.wikimedia.org
>> > https://lists.wikimedia.org/mailman/listinfo/analytics
>> >
>> >
>> >
>> >
>> > _______________________________________________
>> > Analytics mailing list
>> > Analytics@lists.wikimedia.org
>> > https://lists.wikimedia.org/mailman/listinfo/analytics
>> >
>>
>>
>>
>> --
>> Oliver Keyes
>> Research Analyst
>> Wikimedia Foundation
>>
>> _______________________________________________
>> Analytics mailing list
>> Analytics@lists.wikimedia.org
>> https://lists.wikimedia.org/mailman/listinfo/analytics
>
>
>
> _______________________________________________
> Analytics mailing list
> Analytics@lists.wikimedia.org
> https://lists.wikimedia.org/mailman/listinfo/analytics
>
--
Oliver Keyes
Research Analyst
Wikimedia Foundation
_______________________________________________
Analytics mailing list
Analytics@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/analytics
_______________________________________________
Analytics mailing list
Analytics@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/analytics