Database monitoring may crash the server or badly affect its page locking logic [CORE1881] #2313
Submitted by: @dyemanov
The monitoring code unconditionally fetches the header page to report the actual database state. But, if being done at the AST level, this may cause two concurrent pending requests being linked to a single owner. The lock manager highly dislikes this situation and terminates the process via a bugcheck.
It can be reproduced under high load with the SuperClassic architecture. So far I was unable to see this issue on v2.1 Classic, maybe it could be explained by different synchronization techniques used. But anyway, I mark v2.1.0 as possibly affected and the fix will be committed into both branches.
The text was updated successfully, but these errors were encountered: