Background
The Page Previews feature will be deployed in staged rollouts, beginning with Stage 0: Italian, Russian, Greek, Catalan, Hebrew (see rollout plan). The performance on each Wikipedia will be measured and analyzed.
AC
- T157700 is resolved providing 90% roll out
- Restbase is enabled on stage 0 wikis (T158221)
- Determine and apply appropriate sampling rate (0.05%) rOMWC87647a5f7922: Re-enable Page Previews instrumentation
- Update the sampling rates on the schema page too: https://meta.wikimedia.org/wiki/Schema_talk:Popups#Sampling (see https://meta.wikimedia.org/w/index.php?title=Schema_talk%3APopups&type=revision&diff=16375450&oldid=16297000)
- Popups schema is updated to include whether the feature is enabled (isEnabled)
After the beta feature is graduated make sure:
- Those logged in users who enabled the beta feature will still have the feature enabled.
- 90% of logged out users will have the feature delivered to them.
- Events are logged for all users (as a basis of comparison)
- Feature is not present under the beta feature list
- Olga, Anthony and TSG run through test plan
Open Questions
Should this be implemented by making the Page Previews codebase lazily-loadable?No. Given that all of the interaction handling code and the instrumentation will be required, we (the Reading Web software engineers) consider delaying the initial preview too much of a penalty.Should we test all Wikipedias or only a portion?We'll test on all stage 0 Wikipedias.
The original wording of this task, which is as follows, is still relevant but approached the problem from a performance perspective:
There are a handful of instances where the majority of the Popups code needn't be loaded but is:
- When the user has enabled the beta feature but has since disabled it via the settings menu
- When the user is entered into the A/B test but isn't assigned the enabled condition (see T132604)
- When the user has the NavPopups gadget enabled
AC
Answers on the back of a postcard, sent to mobile-l/wikitech-l:
- Can we really defer the loading of the majority of the codebase?
- Would it be worth it, i.e. what the impact on UX, if any, be acceptable?
- If it wouldn't be worth it, are there any small improvements that we can make to improve the status quo?