After container upgrade Horizon crashes during login: key XXXX is missing from offline manifest

[Ivan Kolodyazhny]

The issue is reproduced on horizon upgrade in a containerized environment with memcached. We've got pre-built horizon images with static (manage.py collectstatic && manage.py compress were executed during image build procedure).

During 'manage.py compress' call django-compressor sends a signal to horizon. After signal is received horizon tries to clear cache for templates. In our case, we don't call 'manage.py compress' during deployment, so django-compress tries to read old data from the cache and fails to render a template. There are some logs: [1]

The root cause is:

[1] http://paste.openstack.org/show/792652/

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/721674
Committed: https://git.openstack.org/cgit/openstack/horizon/commit/?id=0d290b9184a554cda7f6dc28d82401bb85cad7b3
Submitter: Zuul
Branch: master

commit 0d290b9184a554cda7f6dc28d82401bb85cad7b3
Author: Ivan Kolodyazhny <email address hidden>
Date: Tue Apr 21 18:36:35 2020 +0300

    Fix django-compress caching issues

    django-compress doesn't work well with cache. If horizon is
    deployed in a containarized environment with pre-built static
    in images, django-compress tries to load old templates from cache.

    This patch introduces following changes:

    * update_angular_template_hash uses the correct Django cache backend now
    * fixed cache cleanup for memcached backend
    * added clean_angular_template_cache management command to clean cache
      manually without executing heavy 'compress' command.

    Closes-Bug: #1874657
    Change-Id: If1fd61fa2e4b36e22797b0f54ac067a2a937f369