Page MenuHomePhabricator

Analyze impact of parallel loading of page content and mobile visual editor code
Open, Needs TriagePublic

Event Timeline

nshahquinn-wmf created this task.

@Esanders, @ppelberg and I talked this week and thought it would be good if you can check into this on your own using your EventLogging access, so I can focus on T212137 and T211239 (and potentially T209705). As you probably know, it would be a relatively simple query, something like:

select
    date_format(dt, "YYYY-mm-dd") as day,
    percentile_approx(event.ready_timing, 0.1) as 10th_percentile,
    percentile_approx(event.ready_timing, 0.5) as 50th_percentile,
    percentile_approx(event.ready_timing, 0.9) as 90th_percentile
event.ready_timing
from event_sanitized.edit_attempt_step
where
    event.platform = "phone" and
    event.editor_interface = "visualeditor" and
    year = 2019 -- We need to specify at least one partition
group by date_format(dt, "YYYY-mm-dd")

Would that work? I'm happy to give more advice as needed 😁

Esanders added a comment.EditedMar 6 2019, 9:32 PM

Thanks, fixed typos:

select
    date_format(dt, "YYYY-MM-dd") as day,
    percentile_approx(event.ready_timing, 0.1) as 10th_percentile,
    percentile_approx(event.ready_timing, 0.5) as 50th_percentile,
    percentile_approx(event.ready_timing, 0.9) as 90th_percentile,
    count(event.ready_timing) as sample_size
from event_sanitized.editattemptstep
where
    event.platform = "phone" and
    event.editor_interface = "visualeditor" and
    year = 2019 -- We need to specify at least one partition
group by date_format(dt, "YYYY-MM-dd")

Here's the 10th/50th percentiles for 2019 (90th looks similar). The data is very noisy with no obvious improvements. :/ Daily sample sizes are <1000

Esanders added a comment.EditedMar 6 2019, 9:49 PM

If you wanted to read the tea leaves you could say there was a slowdown on the Friday after this release:

https://www.mediawiki.org/wiki/MediaWiki_1.33/wmf.16

There are a lot of MobileFrontend changes in that release, and a change to VE mobile tracking:
https://gerrit.wikimedia.org/r/#/c/mediawiki/extensions/VisualEditor/+/487317/

The change we are analysing was in wmf14 which went out on Jan 22 (so 24-25th before we would see impact). There's not much movement beyond noise around those days.

Esanders added a comment.EditedMar 6 2019, 9:58 PM

Just did a test on live and something weird is happening:

More investigation needed...

edit: Filed as T217826

nshahquinn-wmf moved this task from Triage to Tracking on the Product-Analytics board.

Here's the 10th/50th percentiles for 2019 (90th looks similar). The data is very noisy with no obvious improvements. :/ Daily sample sizes are <1000

Nice work! I think the mobile VE oversampling should help deal with the noisiness.