| JoeWalsh | |
| Jul 13 2020, 4:49 PM |
| F31947395: Screen Shot 2020-07-24 at 5.00.02 PM.png | |
| Jul 25 2020, 12:02 AM |
The apps have been updated to send pageview=1 in the X-Analytics header with requests that are pageviews. To exclude requests that are made for offline viewing, the definition should be updated to only count requests with pageview=1 in the X-Analytics header.
iOS App Version with this change:
6.6.2 (released 7/13/2020)
Android App Version with this change:
2.7.50324 (with the header on all mobile-html requests)
TBD (with the header only on pageviews)
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | None | T256508 API pageview counts for 'Mobile app' are incorrect since switch to mobile-html | |||
| Resolved | • Nuria | T257860 Update PageviewDefinition to only include /api/rest_v1/page/mobile-html requests with X-Analytics: pageview=1 in pageviews |
We think it will be very error prone to try and parse out the app version and split up the code to handle different versions in different ways. We think it might be better to only consider pageview=1 requests as pageviews going forward after some point. We should talk and figure out if this would work for you and what point we want to make the switch. As well as discuss alternatives.
Also, who on your team can we work with to code review these changes and test the implications? We usually make the changes and then run them on a few hours of data to see what they mean.
Sounds good - @Charlotte, @JMinor, and @SNowick_WMF can help determine when we should switch to only counting pageview=1 requests. It might be feasible to make this switch fairly soon instead of trying to separate out by version. For code review, our tech leads @Tsevener (iOS) and @Dbrant (Android) can help
Note: @Charlotte, @JMinor, and @SNowick_WMF agree with switching to only counting pageview=1 requests. Heads up @Tsevener (iOS) and @Dbrant (Android).
So we are all on the same page this will *reduce* the number of pageviews, see plot. For 2020/07/24 data for IOS the reduction is about 8% on data marked now as 'user'
So we are all on the same page this will *reduce* the number of pageviews, see plot. For 2020/07/24 data for IOS the reduction is about 8% on data marked now as 'user'
Confirmed, that's expected given that we're now excluding requests to save for offline viewing and requests from versions older than 6.6.2.
Change 616591 had a related patch set uploaded (by Nuria; owner: Nuria):
[analytics/refinery/source@master] Removing outdated IOS pageview code
Change 616629 had a related patch set uploaded (by Nuria; owner: Nuria):
[analytics/refinery/source@master] For Android and iOS we only count pageviews with x-Analytics marker
Current patch counts only requests with X-Analytics: pageview=1 for iOS and Android, logic should be unchnaged for KAiOS
Change 616591 merged by jenkins-bot:
[analytics/refinery/source@master] Remove outdated IOS pageview code
Run 1 hour with new code:
OLD code:
_c0 access_method year month day hour 12789534 desktop 2020 7 27 1 13547212 mobile web 2020 7 27 1 372909 mobile app 2020 7 27 1
As you can see pageviews for mobile app are 89% of what they were before
NEW CODE:
_c0 access_method year month day hour 12789534 desktop 2020 7 27 1 13547212 mobile web 2020 7 27 1 333627 mobile app 2020 7 27 1
Change 616629 merged by jenkins-bot:
[analytics/refinery/source@master] For Android and iOS we only count pageviews with x-Analytics marker
Code is deployed and in effect, there are every hour about 5000 pageviews for mobile app that get marked as such despite not having the pageview marker on X-analytics, will investigate a bit more cause that suggests a bug in the pageview definition. This 5000 amount to 1.5% of total pageviews every hour so a small amount but still it needs some investigation.
Change 618157 had a related patch set uploaded (by Nuria; owner: Nuria):
[analytics/refinery/source@master] Test case for pageviews marked as such that should not be so
Change 618635 had a related patch set uploaded (by Nuria; owner: Nuria):
[analytics/refinery/source@master] Requests with app user agents should not be evaluated as app pageviews
Change 618157 abandoned by Nuria:
[analytics/refinery/source@master] Test case for pageviews marked as such that should not be so
Reason:
Known problems list updated with larger problem: https://wikitech.wikimedia.org/wiki/Analytics/Data_Lake/Traffic/Pageview_hourly#Changes_and_known_problems_since_2015-06-16
Pageviews as of now for 1 hr
select count(*), access_method, year, month, day, hour from wmf.webrequest where is_pageview and year =2020 and day=6 and month=8 and hour=2 group by access_method, year, month, day, hour;
_c0 access_method year month day hour
12300461 desktop 2020 8 6 2
293564 mobile app 2020 8 6 2
12440620 mobile web 2020 8 6 2
With correction on change https://gerrit.wikimedia.org/r/618635 pageviews for mobile app are about 1% lower
OK
_c0 access_method year month day hour
12300461 desktop 2020 8 6 2
12440620 mobile web 2020 8 6 2
290383 mobile app 2020 8 6 2
Change 620293 had a related patch set uploaded (by Fdans; owner: Fdans):
[operations/puppet@production] modules/refine: bump jar version to fix pageview definition bug
Change 620293 merged by Elukey:
[operations/puppet@production] profile::analytics::refinery::job::refine: bump jar version
Mentioned in SAL (#wikimedia-analytics) [2020-08-14T09:13:36Z] <fdans> restarting refine to apply T257860