Change Details

[[ https://superset.wikimedia.org/sqllab/?savedQueryId=1176 | This query ]] gives surprising results: ``` select count(DISTINCT page_title) from pageview_hourly where year = 2025 and month = 9 and day = 30 and hour = 23 and page_id = 12412140 ------ 1336 ``` In this dataset, the same `page_id` can have different titles if these titles redirect to this `page_id`. But in this case looking at the distinct titles shows completely unrelated pages. And this is only true this hour, in other hours this `page_id` only has a handful of distinct titles, usually redirects. This value is [[ https://gerrit.wikimedia.org/g/mediawiki/extensions/WikimediaEvents/+/59f724559d86779507b62221cd6e11b39466b690/includes/WikimediaEventsHooks.php#152 | set here ]] by the WikimediaEvents extension. The way [[ https://gerrit.wikimedia.org/g/mediawiki/extensions/XAnalytics/+/d417192bc6943ea01c91f80b55781d68be27429a/includes/XAnalytics.php#26 | that hook is run ]] seems to say it should be reliable. Something's wrong somewhere, the same X-Analytics header is being sent in response to different requests somehow. Can this be cached / accidentally sent somehow?

[[ https://superset.wikimedia.org/sqllab/?savedQueryId=1176 | This query ]] gives surprising results: ``` select count(DISTINCT page_title) from pageview_hourly where year = 2025 and month = 9 and day = 30 and hour = 23 and page_id = 12412140 ------ 1336 ``` In this dataset, the same `page_id` can have different titles if these titles redirect to this `page_id`. But in this case looking at the distinct titles shows completely unrelated pages. And this is only true this hour, in other hours this `page_id` only has a handful of distinct titles, usually redirects. This value is [[ https://gerrit.wikimedia.org/g/mediawiki/extensions/WikimediaEvents/+/59f724559d86779507b62221cd6e11b39466b690/includes/WikimediaEventsHooks.php#152 | set here ]] by the WikimediaEvents extension. The way [[ https://gerrit.wikimedia.org/g/mediawiki/extensions/XAnalytics/+/d417192bc6943ea01c91f80b55781d68be27429a/includes/XAnalytics.php#26 | that hook is run ]] seems to say it should be reliable. Something's wrong somewhere, the same X-Analytics header is being sent in response to different requests somehow. Can this be cached / accidentally sent somehow? [[ https://wikimedia.slack.com/archives/CSV483812/p1761770840546649 | Slack thread context ]]

[[ https://superset.wikimedia.org/sqllab/?savedQueryId=1176 | This query ]] gives surprising results: ``` select count(DISTINCT page_title) from pageview_hourly where year = 2025 and month = 9 and day = 30 and hour = 23 and page_id = 12412140 ------ 1336 ``` In this dataset, the same `page_id` can have different titles if these titles redirect to this `page_id`. But in this case looking at the distinct titles shows completely unrelated pages. And this is only true this hour, in other hours this `page_id` only has a handful of distinct titles, usually redirects. This value is [[ https://gerrit.wikimedia.org/g/mediawiki/extensions/WikimediaEvents/+/59f724559d86779507b62221cd6e11b39466b690/includes/WikimediaEventsHooks.php#152 | set here ]] by the WikimediaEvents extension. The way [[ https://gerrit.wikimedia.org/g/mediawiki/extensions/XAnalytics/+/d417192bc6943ea01c91f80b55781d68be27429a/includes/XAnalytics.php#26 | that hook is run ]] seems to say it should be reliable. Something's wrong somewhere, the same X-Analytics header is being sent in response to different requests somehow. Can this be cached / accidentally sent somehow? [[ https://wikimedia.slack.com/archives/CSV483812/p1761770840546649 | Slack thread context ]]