On Wed, 2006-05-24 at 00:15, Bruce Momjian wrote: > Jim C. Nasby wrote: > > On Tue, May 23, 2006 at 10:19:25AM +0200, Csaba Nagy wrote: > > > On Mon, 2006-05-22 at 23:55, Jim C. Nasby wrote: > > > > BTW, there's a bug/issue with CLUSTER that makes it not entirely > > > > transaction safe. > > > > > > For God's sake, don't fix that one ! I rely on it... now seriously, > > > until there's a way to tell the DB that an exclusive lock is enough to > > > vacuum all dead rows, even if they would be visible by other > > > transactions, CLUSTER is the only thing I can use to circumvent the long > > > running transaction syndrome. > > > > I believe it's already fixed in HEAD/8.2. > > If it is, TODO doesn't reflect that: > > * Make CLUSTER preserve recently-dead tuples per MVCC > requirements Can I ask to leave in some kind of switch (configuration, syntax, I don't care) so that the old behavior is still available if needed ? Thanks, Csaba.
