Tom Lane wrote:
>
> ................ So, as soon as any backend
> checks a tuple and sees that its inserting transaction did commit,
> it rewrites the tuple with a new state "INSERT KNOWN COMMITTED" (which
> is represented by inserting XID = 0 or some such). .........
>
The way concurrency is supported in PostgreSQL is really cool, and I
think not widely understood. The tuple uses flags stored in the
t_infomask field of the HeapTupleHeader structure to 'cache' the status
of the creating and deleting transactions for each tuple.
Check out backend/utils/time/tqual.c and include/utils/tqual.h for
the details of the algorithms. (Not recommended if you have been
drinking at all)
Ullman "Principles of Database and Knowledge-Base Systems, Vol 1" Has a
pretty good discussion of time based and lock based schemes for
concurrency control.
Bernie Frankpitt
