Re: Low hanging fruit in lazy-XID-assignment patch?

"Josh Berkus" <josh(at)agliodbs(dot)com> writes:

> No, that would be REPEATABLE READ. That's one of the areas where we need to
> test HOT; does RR and SERIALIZABLE still work correctly with HOT?

I'm a little confused by your question. REPEATABLE READ is an isolation level
we don't directly support in Postgres. If you set the isolation level to
REPEATABLE READ you get SERIALIZABLE which is a higher level.

HOT works fine with SERIALIZABLE, It uses the same criteria as vacuum for
determining when a tuple is dead, namely it compares against RecentGlobalXmin.

The check for whether a transaction can see any old tuples in "broken" chains
uses the serializable snapshot as a conservative proxy for the oldest snapshot
which might be in use. That will work for both serializable and read
committed.

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com