When migrating from python 2 to 3 we can get in a situation where the
cache has data that the new version of python can't read. In this case
drop the bad data and re-cache the correct data.
This commit upgrades the version of pymongo from 2.x to 3.x, removing usages to deprecated functions usage and fixing tests where necessary.
This version of pymongo supports MongoDB 2.x all the way up to 4.2, and this ensures that the platform will be able to run on a supported MongoDB version in the next release.
Scheduled emails show "unsubscribe" link if waffle switch
`schedules.course_update_show_unsubscribe` is enabled, and
settings.ACE_ENABLED_POLICIES respects `bulk_email_optout`.
API endpoint allows GET/POST requests, which:
* GET asks for confirmation of opt-out
* POST accepts "unsubscribe" or "cancel", where "unsubscribe" creates the
Optout entry, and "cancel" does nothing.
Fixes flaky tests:
* The resolvers handle users in "bins", which are groups that depend on the user ID.
* The test user ID varies depending on the test order.
* This change ensures that the bin requested matches the user for the test.
The previous behaviour of using a variable causes Django to complain
about changes not reflected in a migration when the default course
mode slug is changed.
The toggle UPDATE_LOGIN_USER_ERROR_STATUS_CODE was added to roll out a
breaking change for `login_user` auth errors to return a 400 rather than
a 200.
This toggle was enabled in Production on 12/5/2019 with seemingly no
adverse affects.
ARCH-1253
We use pickling in conjunction with caches in these cases so we should
be able to fallback to the uncached behavior if the pickled data is
corrupt in some way.
eg. it's a python 2 pickle we don't know how to read.
When going between python 2 and python 3.5 we can get pickeld course
structires that are incompatible no matter what we do do to the bug
linked in the comment. In this case, handle the error and delete the
corrupt data from the cache.
Making this fairly generic because if we have any bad data in cache we
don't want it to blow up the whole process. Just delete the bad data
and try again.
Returning JSON from `login_user` for third party auth failures makes
the response more consistent with all other `login_user` responses.
The only calls to `login_user` with this failure are processed by
`shim_student_view` which will in-turn remove this JSON. This improves
the `login_user` response in advance of switching the logistration page
to use `login_user` without `shim_student_view`.
ARCH-1253
