I recently had this problem on Azure. Created a SQL Warehouse with a 20 minute auto stop setting. When I saw a $1500 bill on my cc in December, I checked on Azure and saw that the SQL Warehouse was up and running since October 8th. I tried to shut it down manually and it just kept running again. I doubled checked the auto stop configuration setting and it was still twenty minutes.

I looked under the query history and I saw that every few seconds  a USE command was failing trying to connect to a database that had been deleted some time ago. I think that somehow this error was preventing the inactivity clock from ticking down. I had to actually delete the SQL Warehouse to make it stop.

I only use the Databricks instance for study and had barely been on it in months.

The damage is an $2800 bill that I am still struggling with Microsoft support on because they don't seem to understand how Databricks costs are supposed to work. If or when I resolve this with them I'll get back on here. I guess I could try to reproduce it, but it should be obvious to them that Databricks and VM costs shot up together, and back down together as soon as I deleted the SQL Warehouse. They seem to think I allocated a VM service, and my VM costs are from that. Will try with them, again tomorrow.