Received: from bond.goldwiretech.com (IDENT:root@goldwiretech.natural.com [206.217.52.129]) by hub.org (8.9.3/8.9.3) with ESMTP id JAA49541 for ; Thu, 15 Jul 1999 09:33:15 -0400 (EDT) (envelope-from rwhit@goldwiretech.com) Received: from goldwiretech.com (shaken [10.0.0.103]) by bond.goldwiretech.com (8.8.7/8.8.7) with ESMTP id JAA07007 for ; Thu, 15 Jul 1999 09:28:20 -0400 Message-ID: <378DE32C.4C3BE7C2@goldwiretech.com> Date: Thu, 15 Jul 1999 09:33:32 -0400 From: Robin Whitworth X-Mailer: Mozilla 4.04 [en] (WinNT; U) MIME-Version: 1.0 To: pgsql-interfaces@hub.org Subject: Re: resolving locking conflicts Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit An alternative strategy is to use logical locks, rather than physical locks. To do this, add a column to your table to indicate whether or not a given record is locked. You might, for example, leave the column NULL when the record is not locked, and set it to the id of the locking user when it is locked. Then, to lock a record do something like: BEGIN; SELECT locked, ... FROM table WHERE ...; if (locked is not NULL) { Error("row locked by user %s\n", locked) } else { UPDATE table SET locked = ; } COMMIT; You can expand on this by adding a timestamp to indicate when the lock was created and allow it to be broken if held "too long". -rwhit > Date: Wed, 14 Jul 1999 13:04:06 +0100 > From: "Burgess, Trevor - HMS" > Subject: resolving locking conflicts > > I am using Postgres 6.5 on Linux through the libpq interface > > I have a situation where an existing record can be opened by one user, > updated or rolled back. While the first user has the record I need to make > sure no other user can open it. I use > > SELECT .... FOR UPDATE > > This works but the second process blocks (looking to the user like a hung > program), until the first process releases the record. Since updating a > record can take some time the blocking could be there for some time. > > I would like the second SELECT .... FOR UPDATE to fail with an error I can > catch telling me of a conflict (and hopefully which process has the lock), > allowing me to backout gracefully with a consolatory message to the user. I > would like normal SELECTs (without FOR UPDATE) to work an normal. > > I think that this is an option in pg_options but I'm not sure. I have looked > in the source (lock.c and lmgr.c) to try and figure out what to do. I found > something called "user locks" which looks promising but I'm still not sure. > > Help ! > > Trevor