Thanks so much. flush() was exactly what I needed. I wasn't using flush() because I was creating my session with autoflush=True, and I was unclear on the semantics of that flag. I thought autoflush=True meant it flushed everything right away, but it doesn't actually flush() until required to do so by a subsequent database access.
-matt On Dec 28, 5:46 pm, Michael Bayer <[EMAIL PROTECTED]> wrote: > Im assuming you're writing an ORM centric application for my answers > below. If not, the answers would be slightly different. > > On Dec 28, 2007, at 8:02 PM, Glypho Phobet wrote: > > > > > Problem #1: I can't get the id of row1 until I commit the > > transaction. I'd like to be able to get, before I commit, the id that > > row1 will have after the commit, so that I can enter it in row2. Or > > is there some way of telling SQLAlchemy that a particular column in > > row2 should point to whatever the id of of row1 will be when the > > commit happens? > > why cant you get the id until commit happens ? with the ORM you just > issue a flush() anytime you want and it will insert records/get new > ids. im assuming you are also using session.begin() and commit() to > frame the larger transaction; within those, you can issue as many > flush() calls as you like and they participate in the same > transaction....so you can do any number of persists, loads, deletes, > whatever, without ever having to commit anything. you can also issue > SQL if you wanted to execute postgres sequences or something like that. > > > Problem #2: If I commit before creating row2, in order to get the id > > of row1, any changes I make to row1 after I commit are immediately > > reflected in the database -- regardless of whether I commit or > > rollback at the end. If I try to re-save row1 after making changes, I > > get a traceback saying that row1 is "already persistent," like this: > > > <class 'sqlalchemy.exceptions.InvalidRequestError'>: Instance > > '[EMAIL PROTECTED]' is already persistent > > > Is there a way to make this object non-persistent again? I want to be > > able to rollback all changes made to row1 and row2 together, if > > there's an error. > > if a transaction fails, you need to remove (or repair, if thats > feasable) whatever offending objects are present in the session. if > your process is going to re-do everything and recreate objects, you > need to clear the whole session using session.clear()..this is > typically the best approach after a transaction fails. usually > within a web application a failed transaction means youre going to > report an error and end the request..if you are "retrying" wihtin one > request and doing everything again (which is unusual), just do a > clear(). > > > Problem #3: Even if I do this in two separate transactions, and > > re-query for row1 by id the second time I need it, SQLAlchemy returns > > an instance that is "already persistent." Probably this is some sort > > of > > caching that SQLAlchemy is doing, which I will be very grateful in > > general, > > but is there a way to turn it off for one query? > > the session always returns the same instance for a particular primary > key once its loaded or persisted. if you want to remove that > instance, use session.expunge(theinstance). or as above > session.clear() to clear the whole thing. --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "sqlalchemy" group. To post to this group, send email to sqlalchemy@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sqlalchemy?hl=en -~----------~----~----~----~------~----~------~--~---
