Roughly 2,500 events per day are missing the required field editingSessionId, which causes the events to be thrown out. I don't see an obvious pattern in which actions exhibit this problem.
Description
Description
Details
Details
| Subject | Repo | Branch | Lines +/- | |
|---|---|---|---|---|
| Change generation of logging session ID | mediawiki/extensions/DiscussionTools | master | +10 -3 |
Related Objects
Related Objects
Event Timeline
Comment Actions
There does seem to be a useful pattern: "integration": "discussiontools" was present on every event I checked. I suspect that this means there's a path to a VisualEditorFeatureUse event firing before the first EditAttemptStep init event, which is normally what would generate the sessionid.
Comment Actions
Change 666451 had a related patch set uploaded (by DLynch; owner: DLynch):
[mediawiki/extensions/DiscussionTools@master] Change generation of logging session ID
Comment Actions
Change 666451 merged by jenkins-bot:
[mediawiki/extensions/DiscussionTools@master] Change generation of logging session ID
Comment Actions
I should note that I couldn't actually reproduce triggering a VisualEditorFeatureUse event which didn't have a session id, so this being the cause is largely speculation. Still, absent some really weird case I think this should fix any possible route to such an event being generated.
Comment Actions
Checking the logstash search, these errors appear to have stopped as of March 4th -- i.e. as of the last train deploy to group2.
Comment Actions
FYI, yesterday VisualEditorFeatureUse and EditAttemptStep have been migrated to Event Platform (T267353, T267343), so the logstash search for their errors has changed. Use this logstash dashboard.
Validation errors are now also monitor-able in Grafana too.
Comment Actions
Well, the error did disappear before that, and I can't find any for VisualEditorFeatureUse on that new dashboard, so presumably my conclusion holds up...