Re: Concurrent CREATE INDEX, try 2 (was Re: Reducing

Ühel kenal päeval, T, 2005-12-06 kell 16:01, kirjutas Tom Lane:
> Hannu Krosing <hannu@skype.net> writes:
> > 1) run a transaction repeatedly, trying to hit a point of no concurrent
> > transactions,
> 
> In the sort of 24x7 environment that people are arguing this is needed
> for, it's entirely possible that that will *never* succeed.

My OLTP transactions are usually 5-50ms in length. common sense tells
me, that if I disallow new transactions for 100ms, I am more than likely
to have waited for all existing ones to have finished and can do my 1 ms
of "take snapshot + commit" and let all the waiting transactions to
commence.

If the database is running longer transactions, there can be a GUC to
adjust the time CUNCURRENT CREATE INDEX will wait. For example for
trx'es mostly in 0.5-2 sec range the wait could be set to 3 sec.

-------------
Hannu