User Preference table clean-up to prevent system performance degradation

[Jeff_GWC]

We recently had some very poor performance with our sub-production instances. So we opened a case with support and they found that we had over 77k User Preferences with my account having over 3.5k. Once I deleted all of my User Preferences (in the sub-production instances) the performance was night and day. I have since gone through and deleted all user preferences tied to inactive users and the performance seems to be back to normal. 

My question and desire is to create a clean-up job so that we can limit our exposure to this issue. I know how to create the clean-up jobs easily enough, but I am not entirely sure what is safe to delete or clean-up within the User Preferences table. 

Here is the document I have been directed to, but it really only says that over a certain number of records you will experience system degradation, but doesn't go into any details about what is safe to clean up and what to leave alone. 
https://docs.servicenow.com/bundle/sandiego-platform-administration/page/administer/navigation-and-u...

Any help would be greatly appreciated. 

Thank you,

Jeff

[Jim Coyne]

I've asked SN for some clarification on that small number of 10,000 records on that Docs article.  That seems like a ridiculously small number.

I have 660 records in production myself, an instance I rarely work in to be honest.  I mean just the list view options can create a large number of preferences (separate ones for each table for things like column to sort on, direction of sort, whether the filter is pinned or not, etc...).

I think if your job deletes records for any inactive user should at the very least help you out there.  Deleting them for active users will affect their user experience.  You might get a large number on your first execution, but then far less if you run it on a daily or weekly basis.  I'd probably run it daily as it shouldn't affect the platform much, especially off-hours.

[Luiz Lucena]

I do have the same question regarding that number. Is not clear if it is total (counting all user preferences) or just system wide preferences. 

We have around 700 users with backend access and everytime (as you said) sort a list, modify columns, that creates a user preference. Also, we have users around the globe, so, if they switch languages, that is another user preference record. 

I could not find though a relationship to the user preference count to poor performance, wish ServiceNow could be more clear on that.