Re: [GENERAL] Odd transaction timestamp sequence issue

2006-04-21 Thread Tom Lane
Jeff Amiel <[EMAIL PROTECTED]> writes: > I thought: > "Each transaction sees a snapshot (database version) as of its > starttime, no matter what other transactions are doing while it runs" That's a correct statement in SERIALIZABLE mode, but in the default READ COMMITTED mode, it's more complicat

Re: [GENERAL] Odd transaction timestamp sequence issue

2006-04-21 Thread Jeff Amiel
it is done using now() But what I don't understand is how the transaction that started first could 'see' the record that hadn't been changed yet by the initial update to 'COMPLETE'? I thought: "Each transaction sees a snapshot (database version) as of its starttime, no matter what other tr

Re: [GENERAL] Odd transaction timestamp sequence issue

2006-04-21 Thread Martijn van Oosterhout
On Fri, Apr 21, 2006 at 09:43:55AM -0500, Jeff Amiel wrote: > PostgreSQL 8.1.2 on i386-portbld-freebsd6.0, compiled by GCC cc (GCC) > 3.4.4 [FreeBSD] 20050518 > > We have triggers on each of our tables that create audit table entries > on each insert/update/delete. > The audit table (in additio

Re: [GENERAL] Odd transaction timestamp sequence issue

2006-04-21 Thread Tom Lane
Jeff Amiel <[EMAIL PROTECTED]> writes: > For example, for id 210210 we have an audit trail that looks like this... > audit_idrecord_idwhen columnold_val > new_val > ----- --- --- > --- > 1

[GENERAL] Odd transaction timestamp sequence issue

2006-04-21 Thread Jeff Amiel
PostgreSQL 8.1.2 on i386-portbld-freebsd6.0, compiled by GCC cc (GCC) 3.4.4 [FreeBSD] 20050518 We have triggers on each of our tables that create audit table entries on each insert/update/delete. The audit table (in addition to containing information about the change that was made) contains a