Re: Transaction start in pg_stat_activity

"Simon Riggs" <simon@2ndquadrant.com> writes:
> On Mon, 2006-11-20 at 10:58 -0500, Tom Lane wrote:
>> There is no way we are putting a gettimeofday() call into
>> GetSnapshotData.  I thought you were focused on performance??

> My understanding was there was already a gettimeofday() call per
> statement which is displayed in pg_stat_activity. It seems relatively
> straightforward to have another column which is *not* updated for each
> statement when we are in SERIALIZABLE mode and CommandId > 1.

What for?  The proposal already covers transaction start and statement
start, and those are the only two timestamps available (without adding
extra gettimeofday() calls).  What you propose will merely repeat one of
them.

            regards, tom lane