Backend for Historical Manuscripts Indexing

Requirements

• Clone of the architecture
• Git
• Make
• Python 3.6+
• pip
• virtualenvwrapper

Dev Setup

git clone git@gitlab.teklia.com:arkindex/backend.git
cd backend
mkvirtualenv ark -a .
pip install -e .[test]

When the architecture is running locally to provide required services:

arkindex/manage.py migrate
arkindex/manage.py createsuperuser

Local configuration

For development purposes, you can customize the Arkindex settings by adding a YAML file as arkindex/config.yml. This file is not tracked by Git; if it exists, any configuration directive set in this file will be used for exposed settings from settings.py. You can view the full list of settings on the wiki.

Another mean to customize your Arkindex instance is to add a Python file in arkindex/project/local_settings.py. Here you are not limited to exposed settings, and can customize any setting, or even load Python dependencies at boot time. This is not recommended, as your customization may not be available to real-world Arkindex instances.

ImageMagick setup

PDF and image imports in Arkindex will require ImageMagick. Due to its ability to take any computer down if you give it the right parameters (for example, converting a 1000-page PDF file into JPEG files at 30 000 DPI), it has a security policy file. By default, on Ubuntu, PDF conversion is forbidden.

You will need to edit the ImageMagick policy file to get PDF and Image imports to work in Arkindex. The file is located at /etc/ImageMagick-6/policy.xml.

The line that sets the PDF policy is <policy domain="coder" rights="none" pattern="PDF" />. Replace none with read|write for it to work. See this StackOverflow question for more info.

Local image server

Arkindex splits up image URLs in their image server and the image path. For example, a IIIF server at http://iiif.irht.cnrs.fr/iiif/ and an image at /Paris/JJ042/1.jpg would be represented as an ImageServer instance holding one Image. Since Arkindex has a local IIIF server for image uploads and thumbnails, a special instance of ImageServer is required to point to this local server. In local development, this server should be available at https://ark.localhost/iiif. You will therefore need to create an ImageServer via the Django admin or the Django shell with this URL. To set the local server ID, you can add a custom setting in arkindex/config.yml:

local_imageserver_id: 999

Here is how to quickly create the ImageServer using the shell:

backend/arkindex$ ./manage.py shell
>>> from arkindex.images.models import ImageServer
>>> ImageServer.objects.create(id=1, display_name='local', url='https://ark.localhost/iiif')

Note that this local server will only work inside Docker.

User groups

We use a custom group model in arkindex.users.models (not the django.contrib.auth one). In this early version groups do not define any right yet.

Usage

Makefile

At the root of the repository is a Makefile that provides commands for common operations:

• make or make all: Clean and build;
• make base: Create and push the arkindex-base Docker image that is used to build the arkindex-app image;
• make clean: Cleanup the Python package build and cache files;
• make build: Build the arkindex Python package and recreate the arkindex-app:latest without pushing to the GitLab container registry;
• make test-fixtures: Create the unit tests fixtures on a temporary PostgreSQL database and save them to the data.json file used by most Django unit tests.

Django commands

Aside from the usual Django commands, some custom commands are available via manage.py:

• build_fixtures: Create a set of database elements designed for use by unit tests in a fixture (see make test-fixtures);
• from_csv: Import manifests and index files from a CSV list;
• import_annotations: Import index files from a folder into a specific volume;
• import_acts: Import XML surface files and CSV act files;
• delete_corpus: Delete a big corpus using an RQ task;
• reindex: Reindex elements into Solr;
• telegraf: A special command with InfluxDB-compatible output for Grafana statistics.
• move_lines_to_parents: Moves element children to their geographical parents;

See manage.py <command> --help to view more details about a specific command.

Code validation

Once your code appears to be working on a local server, a few checks have to be performed:

• Migrations: Ensure that all migrations have been created by typing ./manage.py makemigrations.
• Unit tests: Run ./manage.py test to perform unit tests.
  • Use ./manage.py test module_name to perform tests on a single module, if you wish to spend less time waiting for all tests to complete.

Linting

We use pre-commit to check the Python source code syntax of this project.

To be efficient, you should run pre-commit before committing (hence the name...).

To do that, run once :

pip install pre-commit
pre-commit install

The linting workflow will now run on modified files before committing, and may fix issues for you.

If you want to run the full workflow on all the files: pre-commit run -a.

Debugging tools

Run pip install ipython django-debug-toolbar django_extensions to install all the available optional dev tools for the backend.

IPython will give you a nicer shell with syntax highlighting, auto reloading and much more via ./manage.py shell.

Django Debug Toolbar provides you with a neat debug sidebar that will help diagnosing slow API endpoints or weird template bugs. Since the Arkindex frontend is completely decoupled from the backend, you will need to browse to an API endpoint to see the debug toolbar.

Django Extensions adds a lot of manage.py commands ; the most important one is ./manage.py shell_plus which runs the usual shell but with all the available models pre-imported. You can add your own imports with the local_settings.py file. Here is an example that imports most of the backend's enums and some special QuerySet features:

SHELL_PLUS_POST_IMPORTS = [
    ('django.db.models', ('Value', )),
    ('django.db.models.functions', '*'),
    ('arkindex.documents.models', (
        'ElementType',
        'Right',
    )),
    ('arkindex.process.models', (
        'DataImportMode',
    )),
    ('arkindex.project.aws', (
        'S3FileStatus',
    ))
]

Asynchronous tasks

We use rq, integrated via django-rq, to run tasks without blocking an API request or causing timeouts. To call them in Python code, you should use the trigger methods in arkindex.project.triggers; those will do some safety checks to make catching some errors easier in dev. The actual tasks are in arkindex.documents.tasks. The following tasks exist:

• Delete a corpus: corpus_delete
• Delete a list of elements: element_trash
• Delete worker results (transcriptions, classifications, etc. of a worker version): worker_results_delete
• Move an element to another parent: move_element
• Create WorkerActivity instances for all elements of a process: intitialize_activity
• Delete a process and its worker activities: process_delete
• Export a corpus to an SQLite database: export_corpus

To run them, use make worker to start a RQ worker. You will need to have Redis running; make slim or make in the architecture will provide it. make in the architecture also provides a RQ worker running in Docker from a binary build.

Process tasks are run in RQ by default (Community Edition). Two RQ workers must be running at the same time to actually run a process with worker activities, so the initialisation task can wait for the worker activity task to finish:

$ manage.py rqworker -v 3 default high & manage.py rqworker -v 3 tasks

Metrics

The application serves metrics for Prometheus under the /metrics prefix. A specific port can be used by setting the PROMETHEUS_METRICS_PORT environment variable, thus separating the application from the metrics API.