Opened 13 months ago

Closed 13 months ago

Last modified 13 months ago

#34466 closed Bug (fixed)

Django 4.2 overwrites user-specified psycopg cursor_factory

Reported by: Anders Kaseorg Owned by: Anders Kaseorg
Component: Database layer (models, ORM) Version: 4.2
Severity: Release blocker Keywords:
Cc: Daniele Varrazzo, Florian Apolloner Triage Stage: Ready for checkin
Has patch: yes Needs documentation: no
Needs tests: no Patch needs improvement: no
Easy pickings: no UI/UX: no

Description

Zulip configures a custom cursor_factory that wraps psycopg2.extensions.cursor to collect timing statistics for logging. But this no longer works in Django 4.2 due to 09ffc5c1212d4ced58b708cbbf3dfbfb77b782ca (#33308) and 0e2649fdf40cedc5be7e2c0e5f7711f315e36b84 (#34255) because DatabaseWrapper.get_new_connection now unconditionally overwrites connection.cursor_factory (even for psycopg2).

The configured cursor_factory is being passed to get_new_connection as conn_params["cursor_factory"]. get_new_connection should leave that alone if it’s set. (And if it’s not, it might also be cleaner for to pass the default cursor_factory via a keyword argument to psycopg[2].connect too, rather than mutating it later.)

Change History (8)

comment:1 by Anders Kaseorg, 13 months ago

Has patch: set
Owner: changed from nobody to Anders Kaseorg
Status: newassigned

comment:2 by Florian Apolloner, 13 months ago

Needs tests: set
Triage Stage: UnreviewedAccepted

Sounds like a legit request, we didn't limit this functionality on purpose but rather by accident.

Assuming we fix this we should also make it more pythonic (and I guess I am at fault for it being like it is now) and not check for is True but only truthiness. Either way, a test ensuring that we do not regress here would be good.

comment:3 by Anders Kaseorg, 13 months ago

Removed is True and added a test.

comment:4 by Anders Kaseorg, 13 months ago

Needs tests: unset

in reply to:  2 comment:5 by Mariusz Felisiak, 13 months ago

Needs documentation: set
Patch needs improvement: set
Severity: NormalRelease blocker

Replying to Florian Apolloner:

Assuming we fix this we should also make it more pythonic (and I guess I am at fault for it being like it is now) and not check for is True but only truthiness.

We use is True intentionally, to avoid errors when passing wrong truthy values, e.g."server_side_binding": os.environ.get("USE_SSB") (where os.environ.get("USE_SSB") returns "False").

comment:6 by Mariusz Felisiak, 13 months ago

Needs documentation: unset
Patch needs improvement: unset
Triage Stage: AcceptedReady for checkin

comment:7 by Mariusz Felisiak <felisiak.mariusz@…>, 13 months ago

Resolution: fixed
Status: assignedclosed

In 73cbb372:

Fixed #34466 -- Reallowed setting cursor_factory in DATABASESoptions on PostgreSQL.

Regression in 09ffc5c1212d4ced58b708cbbf3dfbfb77b782ca.

comment:8 by Mariusz Felisiak <felisiak.mariusz@…>, 13 months ago

In 0bc2bbf:

[4.2.x] Fixed #34466 -- Reallowed setting cursor_factory in DATABASESoptions on PostgreSQL.

Regression in 09ffc5c1212d4ced58b708cbbf3dfbfb77b782ca.

Backport of 73cbb372baa45d1fdafd571e2f430a980831f722 from main

Note: See TracTickets for help on using tickets.
Back to Top