On 26.12.2010 21:40, Kevin Grittner wrote:
> To recap, I've had an open question on the Serializable Wiki page[1]
> since January about how we should handle long-running transactions.
> The algorithm published by Cahill et al requires keeping some
> transaction information in memory for all committed transactions
> which overlapped a still-running transaction. Since we need to keep
> this in shared memory, and the structures must have a finite
> allocation, there's an obvious looming limit, even if the allocation
> is relatively generous.
Looking at the predicate lock splitting, it occurs to me that it's
possible for a non-serializable transaction to be canceled if it needs
to split a predicate lock held by a concurrent serializable transaction,
and you run out of space in the shared memory predicate lock area. Any
chance of upgrading the lock to a relation lock, or killing the
serializable transaction instead?
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com