On Wed, Aug 22, 2001 at 06:26:14PM -0400, Tom Lane wrote:
>
> current_timestamp is defined as the time of the start of the current
> transaction (this is not a bug, it's a feature).
>
> If your program is set up to hold a transaction open while it's idle,
> that could cause stale-looking values of current_timestamp when you
> finally do something in the transaction. Unfortunately I believe that
> "autocommit off" mode in the DBD driver does exactly that.
>
the DBD-Driver does not support explicit begin-transactions. Instead
every commit and rollbackstatement issues implicit a new
begin-statement when autocommit off. (The same has obviously to
happen, when the databasehandle is opened, but it is not documented)
This and your explanation about the meaning of current_timestamp is
the perfect explanation of our bug.
I just created a fix by just issuing a rollback-statement at the
*beginning* of each workingsession. This should start a new
transaction and - I hope - dont loose to much performance. When its
working, I'll contact Edmund Mergl, author of DBD::Pg and try to find
a better solution.
I personally would prefer a solution where one must issue explicit
begin and commit/rollback and each action in between will cause an
error when autocommit->off.
thnx a lot.
peter
--
mag. peter pilsl
phone: +43 676 3574035
fax : +43 676 3546512
email: pilsl@goldfisch.at
sms : pilsl@max.mail.at
pgp-key available
