Here:
https://gist.github.com/1363860
This simple test case works just fine, meaning the problem is somewhere
in my application. As you can see from the code, I tried both a
"standalone" test and through Pyramid WSGI chain checking if pyramid_tm
middleware possibly borks something somewhere, but both cases worked fine.
However, I can't really replicate a test scenario fully because the
problematic part of my application involves several models with
relationships, multiple updates. If I use EXACTLY the same flow for the
savepoint, ie:
sp = transaction.savepoint()
try:
... # Try insert
transaction.commit()
except IntegrityError:
transaction.abort()
... # Try Update
Then the result is:
2011-11-14 12:58:29,689 INFO [sqlalchemy.engine.base.Engine][worker 0]
BEGIN (implicit)
... SELECTs and UPDATEs
2011-11-14 12:58:29,726 INFO [sqlalchemy.engine.base.Engine][worker 0]
SAVEPOINT sa_savepoint_1
2011-11-14 12:58:29,726 INFO [sqlalchemy.engine.base.Engine][worker 0] {}
2011-11-14 12:58:29,727 INFO [sqlalchemy.engine.base.Engine][worker 0]
INSERT INTO user_stats (user_id, tstamp, portal, ads_new, ads_modified,
ads_deleted) VALUES (%(user_id)s, %(tstamp)s, %(portal)s, %(ads_new)s,
%(ads_modified)s, %(ads_deleted)s)
2011-11-14 12:58:29,727 INFO [sqlalchemy.engine.base.Engine][worker 0]
{'user_id': 1, 'ads_new': 0, 'ads_deleted': 0, 'tstamp':
datetime.date(2011, 11, 14), 'ads_modified': 0, 'portal': 'test'}
2011-11-14 12:58:29,728 INFO [sqlalchemy.engine.base.Engine][worker 0]
ROLLBACK TO SAVEPOINT sa_savepoint_1
2011-11-14 12:58:29,728 INFO [sqlalchemy.engine.base.Engine][worker 0] {}
2011-11-14 12:58:29,754 INFO [sqlalchemy.engine.base.Engine][worker 0]
BEGIN (implicit)
2011-11-14 12:58:29,756 INFO [sqlalchemy.engine.base.Engine][worker 0]
UPDATE user_stats SET ads_new=(user_stats.ads_new + %(ads_new_1)s),
ads_modified=(user_stats.ads_modified + %(ads_modified_1)s),
ads_deleted=(user_stats.ads_deleted + %(ads_deleted_1)s) WHERE
user_stats.user_id = %(user_id_1)s AND user_stats.portal = %(portal_1)s
AND user_stats.tstamp = %(tstamp_1)s
2011-11-14 12:58:29,756 INFO [sqlalchemy.engine.base.Engine][worker 0]
{'ads_deleted_1': 0, 'portal_1': 'test', 'ads_new_1': 0, 'tstamp_1':
datetime.date(2011, 11, 14), 'user_id_1': 1, 'ads_modified_1': 1}
2011-11-14 12:58:29,956 ERROR [pyramid_debugtoolbar][worker 0] Exception
at http://localhost:6543/ads/view/28328?category=
...
DetachedInstanceError: Parent instance <AdLand at 0x7ffb900faa90> is not
bound to a Session; lazy load operation of attribute 'ad_base' cannot
proceed
The exception being thrown when I, after the savepoint fail, try to READ
a value from a model read/updated within the outer transaction. The process:
0. enter view
1. session = DBSession()
2. load some models
3. change them with POSTed data (including adding/removing/updating
models through relationship class members)
4. session.flush() changes (possibly redundant because savepoint
unconditionally flushes)
5. make savepoint, try insert or update some statistical models (which
are not used or in any way related with those from the outer
transaction), commit or abort the savepoint
6. read some data from models loaded in step 2 (reading again here
because flush() in step 4 returns new IDs etc... so another phase of
update* is required)
7. session.flush() everything again
8. exit view
*update involving external resources like processing related uploaded
files which names depend on IDs and other data obtained after flush in
step 4
The DetachedInstanceError is thrown at step 6, more precisely:
if not ad.ad_base.main_image_id and len(ad.ad_base.images):
...
where ad is a "main" model, ad.ad_base is a relationship() with parent
(of another class), main_image_id is integer and ad.ad_base.images is a
relationship to yet another model. The DetachedInstance means the 'ad'
being detached, so ad.ad_base is void.
So somehow the transaction of the savepoint messes up with the external
session, even though there is no relationship between those and the
models inserted/updated within the inner transaction. No foreignkey or
relationship() connection.
.oO V Oo.
On 11/14/2011 12:04 PM, Vlad K. wrote:
Sure, I'll whip something up.
Thanks.
.oO V Oo.
On 11/14/2011 06:01 AM, Chris McDonough wrote:
I hate to ask, but would it be possible to concoct a short Python script
that creates a SQLA DBSession (with the ZopeTransactionExtension
enabled) that issues transaction.commit/savepoint/abort commands that
demonstrates the problem being encountered? I'd be guessing if I tried
to help with this without that.
- C
--
You received this message because you are subscribed to the Google Groups
"pylons-discuss" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to
[email protected].
For more options, visit this group at
http://groups.google.com/group/pylons-discuss?hl=en.
