Page MenuHomePhabricator

Some wikidata edits not being reflected on WDQS
Closed, ResolvedPublicBUG REPORT

Assigned To
Authored By
Jheald
Apr 4 2025, 3:52 PM
Referenced Files
F59001787: image.png
Apr 7 2025, 7:56 AM
F59001777: image.png
Apr 7 2025, 7:56 AM
F58981092: image.png
Apr 4 2025, 4:35 PM
F58980941: wdqs_missing_edits.tsv
Apr 4 2025, 4:02 PM

Description

Some edits, made 3 April 2025, that are not showing up on WDQS.
In each case these were edits made with QuickStatements 3.0 to add an additional alias :

  • diff ("St John the Baptist Church, Bishop's Tawton")
  • diff ("St Mary, Totnes")
  • diff ("St Stephen's Church, Higham")
  • diff ("St Andrew's, Beelsby")

This is the case for 14 out of approx 14000 edits of this kind I made yesterday, as revealed by this query https://w.wiki/DhN3 that I was using to suggest edits.

I know there were quite a lot of issues with WDQS updating a few years ago, but I thought all was meant to be sorted out now, with the new pipelines for updating the triplestore ?

Event Timeline

Full results of the said query https://w.wiki/DhN3 as of 15:55 GMT 4 April 2025, giving the 14 items with new aliases that WDQS is not seeing.

Columns are QID -- en Label -- commonscat (P 373) -- aliases (separated by '/')

In each case the item should have an alias matching the commons cat (and therefore should not be appearing in the list).

However, even though the edit has been made, and the alias has been added, WDQS isn't seeing it

Jheald updated the task description. (Show Details)

Thanks for the report, I checked the following edits:

to see if I can find them in the page_change stream with:

select * from mediawiki_page_change_v1 where year = 2025 and month = 4 and day = 3 and revision.rev_id in (2334246536, 2334197707, 2334197893, 2334246808, 2334217493, 2334218135, 2334253457, 2334254046, 2334232475, 2334232905, 2334233178, 2334233648, 2334233840, 2334256663);

and found nothing.

I find only 6 of these 14 edits in the revision-create stream...

The updater is using the mediawiki_page_change_v1 stream to source its update while the new updater improved some aspects of the update process it is not able to detect missing events.

There are unfortunately no consensus on how to improve this at the moment and is being discussed in T120242.

I suspect an outage happened and caused the events to get lost between mediawiki and event-gate.

Looking at the number mediawiki log messages mentioning a failure I see several spikes and one yesterday around the times these edits were made:

image.png (595×3 px, 103 KB)

I will manually reconcile these items next Monday but if a new edit is made to these items they should fix themselves.

I suspect an outage happened and caused the events to get lost between mediawiki and event-gate.

Perhaps just worth flagging that these 14 edits were not contiguous; they were made at different times (albeit all within the same couple of hours), and surrounded by edits that *did* go through.

So it perhaps would be a number of momentary glitches we would be talking about, rather than a single outage.

Mentioned in SAL (#wikimedia-operations) [2025-04-07T07:45:06Z] <dcausse> T391122: reconciled 14 wikidata items (lost EventBus/eventgate events)

I suspect an outage happened and caused the events to get lost between mediawiki and event-gate.

Perhaps just worth flagging that these 14 edits were not contiguous; they were made at different times (albeit all within the same couple of hours), and surrounded by edits that *did* go through.

So it perhaps would be a number of momentary glitches we would be talking about, rather than a single outage.

Indeed, these edits were made between 13:25 and 16:08 UTC and zooming over this period we can see a surge in errors:
10 minutes precision:

image.png (591×3 px, 84 KB)

1 minute precision:
image.png (591×3 px, 89 KB)

dcausse claimed this task.
dcausse moved this task from Incoming to Done on the Discovery-Search (2025.03.22 - 2025.04.11) board.

The items have been reconciled, the root cause is unfortunately not fixed and these issues might happen again if we don't get to an agreement on how to handle missing events or make the event platform more resilient.