Re: Transaction isolation levels

Geert Jansen <geert@boskant.nl> writes:
> For my curiosity, what is the reason that in read comitted mode MVCC can
> guarantee a consistent snapshot durign a query, while in serializable
> mode MVCC can guarantee it for a longer interval (the entire
> transaction). Are these different MVCC implementations, or is some kind
> of locking performed when executing queries in read committed mode?

It's the same implementation in either case; it's just a matter of which
"snapshot" we refer to when deciding whether recently-committed row
versions are visible to our own query.  The snapshot is basically a
list of open transactions, which are to be considered not-committed
even if they in fact commit while our query or transaction continues.

            regards, tom lane