Overall performance of the extension before we make it the default behavior on Chinese, Greek and Catalan wikipedia. It would be nice to get T67845 reviewed too.
Description
Status | Subtype | Assigned | Task | ||
---|---|---|---|---|---|
Declined | None | T88882 Move Hovercards to OOjs and OOjs UI | |||
Resolved | None | T92555 Turn off 2 month test release of Hovercards on Catalan, and Greek Wikipedias on 2015-06-18 | |||
Resolved | • Tnegrin | T88164 Make hovercards default for Catalan and Greek Wikipedias for 2 months starting 2015-04-15 | |||
Resolved | ori | T88170 Overall performance review of Popups/Hovercards extension | |||
Resolved | • werdna | T88167 Implement instrumentation for new schema | |||
Resolved | leila | T88166 Finalize on schema for the new instrumentation | |||
Resolved | Prtksxna | T88165 Use sendBeacon for event logging | |||
Resolved | Prtksxna | T70341 EventLogging dependency isn't so optional |
Event Timeline
Last time I checked (quite a while ago), Hovercards were using uncacheable API calls. I pointed that out at the time but it wasn't a priority for the beta release.
I don't think it would be a problem on the server side load-wise but it would hurt a lot the end-user performance due to RTT, especially for this particular country selection.
Sorry for the delay in completing the performance review. The fault is entirely my own.
It looks OK for initial deployment. I don't think we should transition from opt-in to all page-views on a major wiki like zhwiki without intermediate steps, though, so I recommend restricting the initial deployment to Greek and Catalan wikis.
There does appear to be a blocker, though: https://gerrit.wikimedia.org/r/#/c/197034/ is unmerged.
I just re-checked on Greek Wikipedia. Hovercard AJAX responses come with "Cache-control: private, must-revalidate, max-age=0" (from MediaWiki) which means they are not edge-cached. I mentioned this before above — @ori, any reason you didn't comment on it?
In my opinion, this is not ready to be deployed more widely and I would argue that we're not doing it any justice by having it deployed to languages whose primary audiences are far away from origin/eqiad.
Given @ori comment about https://gerrit.wikimedia.org/r/#/c/197034/ being the only blocker which is now merged.