18

This is related to Stack Exchange is intermittently failing at random areas and, possibly, The list of related questions is shown empty, although the latter one involved all questions and there was at least an empty "Related" header showing. Questions on this site posted before Search on MSE is broken have related items showing, but I've confirmed none of the new questions since then do, including even having a "Related" header. I've also noticed that recent Math SE questions exhibit this same behavior, so the problem appears to be network wide.

The discussion among the comments of Stack Exchange is intermittently failing at random areas indicated the issue was related to searching, with a link to Nick Craver's tweet provided indicating that issue had been fixed. Nonetheless, the timing of having related items no longer being shown (and likely not even being determined) for any new questions at the same time strongly indicates to me this was also an issue that occurred then but which has not yet been fixed.

Update: As Sonic the Masked Werehog's comment stated, this appears to be a regression of a much earlier bug described in "Related questions" section not showing up on new questions.

3
  • Based on the comments on Nick's answer to the previous question, this doesn't have anything to do with the outage. Her comments are two days old.
    – Catija StaffMod
    Commented Aug 10, 2020 at 5:44
  • 6
    Seems like a regression of this much earlier bug. Tagging accordingly. Commented Aug 10, 2020 at 6:15
  • 1
    It's been 3 weeks and this is still happening. It looks like last time it happened the problem was resolved within a day. I wonder if it's possible to have any kind of update? Commented Aug 31, 2020 at 19:43

1 Answer 1

8

Elastic's rebuilding of related questions was causing some issues a little while back and we did some work to stabilize our internal cluster. Fixes have since been deployed but the mechanism and how it inter-relates to new questions changed as well. In short: we had this turned off for old question rebuilds and inadvertently affected new question rebuilds.

What happens when a question is asked, so that we do not block on an elastic query (which isn't the cheapest thing in the world), we effectively queue the question for a rebuild (anything without a rebuild date is fair game). A job in the background crawls questions to rebuild. Unintentionally, the disabling of rebuilds affected not only anything older than n days, but things without any date as well. For context: the n there is a setting on our end, since what a question is related to may change as new questions are added over time.

The cluster is now back from sabbatical and crunching through related questions. We'll also work to separate these two in the future so that us disabling an optimization will be something users probably never even notice if or when we need to do it again.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .