Konstantin Knizhnik <k.knizhnik@postgrespro.ru> writes:
> But let me ask you one more question: why do we obtaining snapshot twice
> in exec_simple_query:
> first for analyze (pg_analyze_and_rewrite) and one for execution
> (PortalStart)?
That would happen anyway if the plan is cached. If we were to throw away
all plan caching and swear a mighty oath that we'll never put it back,
maybe we could build in a design assumption that planning and execution
use identical snapshots. I doubt that would lead to a net win though.
Also note that our whole approach to cache invalidation is based on the
assumption that if session A needs to see the effects of session B,
they will be taking conflicting locks. Otherwise sinval signaling
is not guaranteed to be detected at the necessary times.
regards, tom lane
