User Story:
As an analyst, I want the ability to look at all hover events
Description:
As per the results of [T141922], we want to log all hover events (vs only logging events with totalInteractionTime > 250ms)
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | CKoerner_WMF | T143169 Announce hovercards test to Russian Wikipedia | |||
| Resolved | ovasileva | T136746 Initiate Hovercards A/B Test 2: ru and it | |||
| Resolved | • jhobs | T145379 Record all hover events | |||
| Resolved | • Jhernandez | T145152 Scenario failure in Selenium tests: Popups can be enabled via the "Enable previews" footer link |
Suggested acceptance criteria; When making this change we're likely to see more events. We should determine whether we need to change the sampling rate after this change.
@jhobs you were right, dismissed is after seeing the hovercard 😆 I got confused, sorry about the kickoff chat thing
See
Change 310315 had a related patch set uploaded (by Jhernandez):
Log all dwelledButAbandoned events
Change 310324 had a related patch set uploaded (by Jhernandez):
Update Schema to the last version
@Jdlrobson @bmansurov @jhobs Want to signoff on this (since I worked on it?)
Beta cluster doesn't seem to have popups (:S wat) so I've updated staging, and I've verified myself there too.
mw.trackSubscribe('event', (topic, event) => console.log(topic, event.action, event)) type that in the console and quickly hover over links and see that the total interaction time is less than 250ms and it is logged.
Don't we sign off on tasks after we verify they work (either on beta cluster or reading web staging)?
Yeah, actually, I think you're right. Sorry, I've been thinking too much in a config change world this week! I would still say we might want to leave it Open until it's no longer a blocker for the config change, but I suppose the person SWATting the config could just SWAT any patches associated with this task as well.
Resolving since MW-1.28-release (WMF-deploy-2016-09-20_(1.28.0-wmf.20)) train has started this week. T136746 should only be deployed to 1.28.0-wmf.20.
Question: Did this change affect only the test group, or both the test and the control group? Asking because I was surprised to find that the number of link interactions recorded per pageview is now much higher than in June/July (concretely[1], for anons on huwiki it rose from 1.4 measured earlier to 5.2 now for hovercards OFF, and from 1.0 in June/July to 4.6 now for hovercards ON).
If the switch to recording all hovers was indeed made for both test and control in the same way, that's actually fine (indeed it was what I had in mind when first bringing this up, because it enables direct comparison) - but the task description says otherwise.
# repeat https://phabricator.wikimedia.org/T139319#2476014 for later timespan and new schema version, # restricted to Firefox: SELECT event_popupEnabled, count(DISTINCT event_linkInteractionToken)/count(DISTINCT event_pageToken) AS link_interactions_per_page FROM log.Popups_15906495 WHERE wiki ='huwiki' AND event_isAnon = 1 AND LEFT(timestamp, 8) >= '20160925' AND LEFT(timestamp, 8) < '20161023' AND INSTR(userAgent,'Firefox') AND NOT INSTR(userAgent,'Seamonkey') AND (event_action = 'pageLoaded' OR event_linkInteractionToken IS NOT NULL) GROUP BY event_popupEnabled; +--------------------+----------------------------+ | event_popupEnabled | link_interactions_per_page | +--------------------+----------------------------+ | 0 | 5.2049 | | 1 | 4.5728 | +--------------------+----------------------------+ 2 rows in set (4 min 19.53 sec)
@Tbayer All groups indeed. The title and description are not what happened, I'll update them