- Use gitlab-ci services instead of docker-compose to run postgres and
  redis
- Switch to docker socket

Allow to use a scanner configured as keyboard to register items
via the web UI.
1. The attributes should be selected manually
2. Focus set manually on ICS id
3. Scan an ICS id -> prefix removed and focus set to serial number
4. Scan serial number
5. Click on submit
6. Focus set automatically on ICS id
7. Repeat from 3

Every tagged release is automatically deployed to staging
by launching the job template from Ansible AWX

Let alembic handle the database creation and migration

CI_COMMIT_REF_NAME is the branch or tag name
1. When pushing a tag the TEST_IMAGE doesn't exist. We always pull master
for cache.
2. Make sure the pull of the cache image doesn't fail even if the image
doesn't exist

ResolvePackageNotFound:
  - blinker 1.4 py36_0
  - wheel 0.30.0 py_1

Update flask-bootstrap to 4.0.0.0.dev1

Allow to keep track of batches of small parts.

Command shall be run every night to keep the users in the database in
sync with the LDAP server.
If a user is not found:
- set its groups to []
- revoke all its tokens

The string representation of a Network is the vlan_name.
This is what we should allow to pass as string to create an interface.

Note that if we pass network_id instead of network when creating
and interface, the validate_interfaces method from the Network class
is not called.

The user_id is stored in the flask session. On every page load, we query
the database to get:
- the user objet (user_account table)
- the user's groups (group table) to know what the user can access

If we try to cache the user_load function, the following error is raised
when retrieving the groups:
sqlalchemy.orm.exc.DetachedInstanceError: Parent instance <User at
0x7f51e050a940> is not bound to a Session; lazy load operation of
attribute 'grp' cannot proceed

The groups from AD are only stored to map them with CSENTRY_LDAP_GROUPS.
We don't do any query on them (get all users from one group).
As we always query them with one user, it's more efficient to store them
in the user_account table as an array than in a separate table.
This allows to easily cache the load_user function.

Note that we have to remember to add the current_user to the sqlalchemy session if
we want to modify it or access a relationship (like tokens)!

The load_user function is now cached for 30 minutes. This decreases a
lot the number of database queries.

Fix INFRA-156

All endpoints are plural

Pagination information (next, prev, first, last) is included in the Link
HTTP header.
The total number of entries is provided in the X-Total-Count HTTP
header.

gateway tag shall be admin_only

Only keep the values really required by the application.

Fix failure for ajax request GET /inventory/_retrieve_items?draw=7&columns...:
invalid request block size: 5574 (max 4096)...skip
	# Please enter the commit message for your changes. Lines starting

Sorting on manufacturer/model/location/status requires
a join which removes the row with empty field.
Disable it until we find a way to sort properly including null values.

When there are more than a 1000 items in the table, retrieving and
displaying all items takes about 7 seconds.
This is every time we refresh the items page, which is too slow.

Some tags (like gateway) should only be selectable by admin users.

- The query to retrieve used ICS ids was run at each iteration of the
  loop...
- Use a set to quickly test the existence of used ICS ids

Enable to run the application closer to the production environment

Having the description field only on the Model class makes things
inconsistent.
It's nice to have a description on other attributes as well.

The generation of the QR code image is quite slow and makes the page
displaying them take a long time to load when it includes more a few.
The data never changes and can easily be cached (the repr() of the class
is used in the cache key, so if the name or id changes, the cache will
be recomputed - there is no need for a timeout).