Explicit closing cursors is required for the consistency in multithread environment

[Vsevolod Novikov]

Using django ORM in multithread environment causes rare (ignored) exceptions like 'ProgrammingError: (2014, "Commands out of sync; you can't run this command now")' in Cursor.__del__ method (this particular error happens when using MySQL backend).

I've met this problem using django ORM (over MySQL) together with twisted.thread.deferToThread call.

The problem happens relatively rare, but is critical for 24/7 services, because looks like causing regular memory leaks.

Deep investigating the code concerned to the exception contexts, I've found that almost no one cursor got by django ORM from the backend is closed explicitly.

I've changed the django code to implement explicit cursor closing in all found contexts. After these changes my service leaved for 12 hours of continuous working shown no one exception like this.

All changes have been made in one file, django/db/models/sql/compiler.py

Changes have been made in the following functions and methods:

• SQLCompiler.execute_sql in SINGLE and MULTI cases
• SQLInsertCompiler.execute_sql
• SQLUpdateCompiler.execute_sql
• order_modified_iter
• also not_modified_iter iterator (like order_modified_iter) has been added to catch iterator finalization stage

All changes are explicitly closing the cursor got from the backend after use.

The applied patch has been made for django v.1.3.1.final

[Vsevolod Novikov]

Patch for django 1.3.1.final to close opened cursors explicitly

[Anssi Kääriäinen]

The reasoning in the description seems valid. I guess the error is the garbage collector tries to collect the cursor while another thread is doing some work. Seems like some backend libraries do not like that. I haven't actually tested this, but in any case closing cursors explicitly is recommended, and we should do it.

It would be great if you could provide a self-contained test case. Of course, having a test in Django's test suite would be perfect, but seems hard to achieve.

[Tim Graham]

I think this is a duplicate of #21751 which has a patch to close cursors after they are used.