Robert Haas wrote:
>> ... serializable transaction ...
> If we were to construct a database that had one giant lock for the
> entire database so that only a single query could execute at one time,
> transactions would be serializable (because they'd in fact be
> serialized). However, performance would suck.
I wonder if this giant-lock-for-isolation-level-serializable
is a mode postgres should support. ISTM it would meet the
letter of the spec, and at least some of the people using
"transaction isolation level serializable" are doing so precisely
because they *want* the database to deal with all possible
serialization issues, and accepting performance penalties.