We recently made a change to convert the handle list lock from a spinlock to a read-write lock (WT-3115). It makes sense to convert the table lock into a read-write lock in a similar way, otherwise if a table is opened by many sessions concurrently they will end up contending on the spinlock.
Before embarking on this change we should investigate whether we can get rid of the table lock altogether by switching tables to be dhandles. If that change isn't significantly more work than switching the lock type it is a preferable approach - since we'd like to move in that direction eventually.
- related to
-
WT-3218 unexpected checkpoint ordering failures
- Closed