On Nov 12, 2008, at 3:59 PM, Joshua Bronson wrote:
Hey Luke,
I've got the 0.4.1 release candidate running on dev with the
minibrain in a separate process, updating feeds every half hour. So
far so good, but I'm gonna keep an eye on the logs between now and
the end of the day tomorrow.
Cool.
Here are some notes on the process I kept along the way:
1. In testing locally, I wiped my databases and recreated them. When
I reran paster setup-app, the admin and the anonymous user were not
created. I ended up adding session.commit() in
melk.model.userinfo.put() because the session.flush() alone was not
enough to get the changes to stick. I'm not checking in this change
though because it was a quick fix, and not necessarily the right fix.
Possibly the stuff run in websetup.py is missing a
model.session_complete() (an explicit commit is also possible since
it holds the actual Session object). It looks like it's just closing
the session at the end. UserInfo.put() should not call commit. Like
the rest of it, we don't want to implicitly commit unknown other work
pending inside the transaction.
2. The problem you helped me with this morning where the minibrain
was not calling init_model with a configured sqlalchemy Session.
Resolved by r1757.
3. melk.util.spider._fetch(..) was sometimes causing a TypeError:
'NoneType' object is not iterable. Resolved by r1764.
Cool.
4. After the last three issues were resolved for me on my machine, I
got an environment set up on dev in /usr/local/melk/deploy/0.4.1,
and proceeded to run the only migration that needed to be run on
monolith_big: 16 -> 17. The migration failed with
Duplicate entry '235964-26677' for key 2") 'ALTER TABLE
filter_response ADD UNIQUE (item_id, filter_id)'
so somehow a non-unique score must have snuck in the monolith.
yeah, this was a possibility beforehand right? We'll want to be able
to clear those up before that migration is run against the real
database in case they are there as well.
Should I look into the last issue more or work on my other remaining
0.4.1 tickets for the rest of this afternoon and tomorrow (I'm
looking at you #298)?
I'd say make that migration able to handle potential pre-existing
duplicate scores before other stuff. If you're taking stuff over from
the trunk, I have some small fixes that went in today that should also
head into 0.4.1.
- Luke
