Page MenuHomePhabricator

TwoColConflict "exit" metrics failing
Closed, ResolvedPublic2 Estimated Story PointsBUG REPORT

Description

In T245420: Record "must-have" information about potential data loss in TwoColConflict we added information about the UI state and how exit happened. These are valuable for giving us a more accurate "success" metric, information about abandonment, and clues about data loss caused by design issues. Unfortunately, our logging schema was too complex and breaks some new guidelines, as well as breaking the server software meant to ingest statistics.

Specifically, the "selections" array in this version of the exit schema cannot use "oneOf" unions, cannot be an array, and so on.

We need to rewrite the column edit and selection data to either pack it into a string, or build an arbitrary string:string map. Examples of valid potential formats:

  • selections: "y<n|n_n|y<n"
  • selections: { "line1": "y<n", "line2": "n_n", "line3": "y<n" }

Neither is great for direct querying, but we've already got a post-processing step in our data pipeline, so we might as well go with the a compact and foolproof string format.

Whatever we do here needs to also work for talk pages.

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript
awight changed the subtype of this task from "Task" to "Bug Report".Apr 4 2020, 8:27 AM
awight set the point value for this task to 2.

Pulling this into the sprint, the bug was causing analytics alerts and we're losing important data that will help make upcoming product decisions.

awight updated the task description. (Show Details)

Change 585866 had a related patch set uploaded (by Awight; owner: Awight):
[mediawiki/extensions/TwoColConflict@master] [WIP] Flatten exit logging

https://gerrit.wikimedia.org/r/585866

Change 585971 had a related patch set uploaded (by Awight; owner: Awight):
[mediawiki/extensions/TwoColConflict@master] Urgent fixes to the "exit" schema

https://gerrit.wikimedia.org/r/585971

Let's remove this field and get the basic data coming through, then work on a more refined fix later:
https://meta.wikimedia.org/w/index.php?title=Schema:TwoColConflictExit&diff=19954553&oldid=19928888

Change 585971 merged by jenkins-bot:
[mediawiki/extensions/TwoColConflict@master] Urgent fixes to the "exit" schema

https://gerrit.wikimedia.org/r/585971

Mentioned in SAL (#wikimedia-operations) [2020-04-06T11:53:14Z] <awight@deploy1001> Synchronized php-1.35.0-wmf.26/extensions/TwoColConflict: SWAT: [[gerrit:586309|Backport talk page and EventLogging changes (T248243, T249404) (duration: 00m 59s)

Deployed. Waiting a day or so for Analytics to remove this schema from the blacklist.

Change 585866 abandoned by Awight:
[WIP] Flatten exit logging

Reason:
Moved work to I6e47d3

https://gerrit.wikimedia.org/r/585866