Oracle GIS testing -- fails to destroy old leftover test database after unclean ending

[Shai Berger]

If you run Django tests and they end without proper clean-up (e.g. the test process is forcibly killed) then the next test run will try to destroy the test-database before rebuilding it from scratch.

On Oracle GIS, this destroy-leftover-database process fails, because the backend tries to drop the test tablespace (DROP TABLESPACE TEST_DB INCLUDING CONTENTS), and for some reason the GIS artifacts make this fail.

The workaround is to drop first the test user (DROP USER TEST_USER CASCADE), but currently this needs to be done manually.

[Shai Berger]

PR: ​https://github.com/django/django/pull/3367

[Shai Berger <shai@…>]

In 2e1ed5f33cf5756b52a10ceadd0ce8e9ecf989e3:

Fixed #23649 -- Made Oracle test-db creation drop test-user earlier when needed

Checked database destruction for ORA-29857, which happens when objects of
some types are left over in an old test database (notably, such objects
are created by the GIS backend). When this happens, we can resolve the
issue by dropping the test-user (if we are allowed to).

An alternative approach, to just switch the order of creation (so that,
if the test user exists and can be dropped, it is always dropped before
the tablespace) was considered; but since the user creation depends on
the tablespace, this would necessitate separating the dropping of the user
from its creation -- a change I am reluctant to make.

Thanks Josh Smeaton and Tim Graham for reviews.

[Shai Berger <shai@…>]

In a1578055e61e8815d1b6d445e8b24ad02d38c3ee:

[1.7.x] Fixed #23649 -- Made Oracle test-db creation drop test-user earlier when needed

Checked database destruction for ORA-29857, which happens when objects of
some types are left over in an old test database (notably, such objects
are created by the GIS backend). When this happens, we can resolve the
issue by dropping the test-user (if we are allowed to).

An alternative approach, to just switch the order of creation (so that,
if the test user exists and can be dropped, it is always dropped before
the tablespace) was considered; but since the user creation depends on
the tablespace, this would necessitate separating the dropping of the user
from its creation -- a change I am reluctant to make.

Thanks Josh Smeaton and Tim Graham for reviews.

Backport of 2e1ed5f33c from master

[Tim Graham <timograham@…>]

In 6ce6c77e03fea68ef1bdadd45f15e403f5b71057:

Added 1.7.1 release notes for refs #23649.

[Tim Graham <timograham@…>]

In 55a5f0b49a7adca008076fa0af6f67ef64d575d8:

[1.7.x] Added 1.7.1 release notes for refs #23649.

Backport of 6ce6c77e03 from master

[Shai Berger]

Thanks, Tim, for adding the release notes I neglected.