> I'm confused. I thought that read-only read-committed transactions did not > consume transaction ids. > By nature they don't need the protection of the transactional system. They > don't make changes so > there's no need to roll back. Nor is there any need to prevent others from > reading their uncommitted > changes. They don't offer consistent reads, so they don't need to prevent > garbage collection. Why > do they need unique transaction identifiers? It is how RO RC transactions was implemented at IB6 :) We discussed this long time ago and decided to leave it as is (i.e. to left RO RC transactions to have own numbers) by the following reasons: a) monitoring: transactions naturally identified by transaction_id b) temporary tables: GTT's are writable in RO RC transactions
Regards, Vlad
