You are browsing a read-only backup copy of Wikitech. The live site can be found at wikitech.wikimedia.org
A change to the mediawiki-core api caused EventLogging schemas to be returned with "required": "" instead of "required": true. This caused a schema validation error and stopped all events from validating. While debugging, we lost about 2 hours of data. We can not backfill the lost data.
Tuesday, April 28, 20:52
Noticed that EL was crashing and throwing an error on every event. After help from Bryan Davis and Kunal Mehta we found out this was due to a mediawiki-core change that had just been deployed. They started working on it with Unbreak Now priority (thanks guys).
Tuesday, April 28, 22:44
This patch to core resolved the issue: https://gerrit.wikimedia.org/r/#/c/207297/ and this bug was filed to track the problem from the analytics point of view: https://phabricator.wikimedia.org/T97487.
Tuesday, April 28, 23:03
Event Logging recovered in Icinga. All is well again.
We can't backfill the lost data because we had to stop the processes that create the backup logs in order to debug. A fairly quick fix that's available right now is to start sending events to the kafka pipeline. This would give us a buffer and let us turn off the single point of failure Event Logging server to do maintenance, debugging, etc.
- Status: Done Fix to API so EventLogging schemas are served as expected.