I've found it's occasionally necessary to go one lever deeper and use the _pg module directly so you are in control of your transactions (instead of letting pgdb automatically put you in one).
vincent wehren wrote: > "Christopher J. Bottaro" <[EMAIL PROTECTED]> schrieb im > Newsbeitrag news:[EMAIL PROTECTED] > | Hi, > | Why is there no support for explicit transactions in the DB API? I mean > | like transaction() to start the trans and commit() and rollback() would > end > | the trans or something. > | > | The reason why I ask is because I wrote a daemon that interacts with a > | Postgres DB. The value of CURRENT_TIMESTAMP according to Postgres is NOT > | the actual walltime, but the walltime when the current transaction > started. > | > | This gets weird when using the Python DB API to interact with Postgres > | because a transaction gets started in 3 places: connection, commit, > | rollback. > | > | So consider the daemon: > | > | [pseudo code] > | connect # begin trans at 12:00 > | sleep waiting # lets say 15 mins > | wake up > | put entry in db using CURRENT_TIMESTAMP # oops > | [/code] > | > | Oops, the CURRENT_TIMESTAMP evaluates to 12:00, not 12:15. > | > | Now I know there are ways around this... > | 1) In Postgres, you can get the walltime and cast it to a timestamp. > | 2) In Python, you can just do an empty commit in order to "manually" > start > | a new transaction. > | > | I just think its a bit weird because this bit me in the butt for quite a > | while and this didn't happen when doing the same thing in other langs. > | > | Anyone have any opinions on this? > > The described behavior seems to be totally in synch with expectations > > (both in accordance with PostgreSQL docs and the DB-API.) > > These "other languages" *must* be doing something wrong! ;) > > ( Auto-commit set to "on" perhaps? ) > > > > Regards, > > - > > Vincent Wehren > > > | > > > -- http://mail.python.org/mailman/listinfo/python-list
