Request ID: 070d7a27354a68fc5d4ae980
channel: JobExecutor level: ERROR wiki: sdwiki message: Failed creating job from description c_message: Page *** does not exist job_type: cirrusSearchLinksUpdate
Loss of certain jobs (type "cirrusSearchLinksUpdate" in this case). The failure seems deterministic so both with and without retry, there is no path to recovery for these updates.
I don't know currently whether these updates are meant to work (e.g. they should work but are failing), or whether it is a case of one part of the system trying something another part doesn't support (in which case the only thing we need to do is make it not queue these jobs).
As I understand it, our validation is not significantly different at execution time than at queuing time, so it's unclear why this fails at execution time instead of at queueing time.
Recorded 560 times in WMF Logstash in recent weeks.
Most breakdown factors show a fairly equal distribution, except the wiki ID. It only affects a small subset of wikis, others are entirely unaffected