Skip to content

Latest commit

 

History

History
268 lines (237 loc) · 11.4 KB

README.md

File metadata and controls

268 lines (237 loc) · 11.4 KB
Raw

Last updated: June 16, 2022

What is this?

reports is Django project for compiling dynamic reports using R Markdown and KoBoToolbox. It is built on the R package knitr: http://yihui.name/knitr/.

This code currently runs on https://data.equitytool.org and relies on https://kf.kobotoolbox.org for authentication, form deployment, and submission collection/storage.

Administrative Reports

To create an administrative report that lists all the users and projects stored in the database:

  1. Log into https://data.equitytool.org/admin/ as a superuser;
  2. Click "+ Add" next to "Admin stats report tasks";
  3. Click "SAVE";
  4. A screen listing "new report" first as well as previous reports below will appear;
  5. Refresh this page every few minutes until "new report" changes into "complete report";
  6. Once that happens, click "complete report";
  7. Finally, click "equitytool_admin_stats....zip" (to the right of "Result") to download the ZIP file containing the statistics.

Application Structure (Django Models)

equitytool.Form

An administrator-defined form that regular users can select when they create new projects. The form is then sent to the regular user's linked KoBoToolbox account, creating a new project that, in turn, is referenced by a reporter.Rendering

reporter.Template

These R Markdown templates transform collected data from its raw state into formatted reports (narrative, tables, charts). They are available read-only to all users, but may only be changed by superusers via the https://data.equitytool.org/admin/ (Django Admin) interface.

reporter.Rendering

Connects the R Markdown reporter.Templates to data collected with KoBoToolbox. There is one instance of this for each user-created project.

reporter.UserExternalApiToken / User Authentication

This application uses a special interface of kf.kobotoolbox.org to create KoBoToolbox users without email confirmation. Once a user registers in this way, they are then authenticated by sending their credentials over HTTPS to the KoBoToolbox server (see reporter.KoboApiAuthBackend). If the credentials are correct, KoBoToolbox returns an API key, which this application then stores in reporter.UserExternalApiToken. That key then authenticates subsequent requests to deploy forms and retrieve submissions.

There are also local-only users whose passwords (hashed) are stored directly in this application's database. Superusers are an example of this. These users have privileges to administer this application but not the connected KoBoToolbox instance. Local-only users also cannot create data collection projects as they have no access to KoBoToolbox.

equitytool.AdminStatsReportTask

This model allows administrative reports to be generated in the background where they are not subject to the same time limits as ordinary web application requests.

R Markdown Templates

Mustache tags are supported so we can return a warning message if a deployment has fewer than 150 responses. See reporter.tests.TestRendering.test_warning.

Variables Available in Templates

  • rendering__name: Rendering.name, i.e. the user's name for the project
  • form__name: Form.name, e.g. Tajikistan (DHS 2012)
  • request__show_urban: the value of ?show_urban= in the URL used to request the report

Known Issues

  • Tables in DOCX exports do not appear properly in LibreOffice: see jgm/pandoc#515

Production installation with Dokku

For this example, we're using an AWS t3.small EC2 instance, which provides 2 GiB of RAM, along with a 40 GiB EBS root volume. It is running Ubuntu 20.04.

  1. Due to the limited amount of RAM, add a 5 GiB swap file: 
    sudo fallocate --length 5G /swapfile
sudo mkswap /swapfile
sudo chmod 600 /swapfile
    Add the following to /etc/fstab to enable swap at each boot: 
    /swapfile none swap sw 0 0
    Enable the swap now: 
    sudo swapon -a
  2. Install Dokku: 
    wget https://raw.githubusercontent.com/dokku/dokku/v0.24.7/bootstrap.sh
sudo DOKKU_TAG=v0.24.7 bash bootstrap.sh
sudo reboot
  3. Create the Dokku application: 
    dokku apps:create data.equitytool.org
  4. Configure the new application: 
    dokku config:set data.equitytool.org ALLOWED_HOSTS='data.equitytool.org'
dokku config:set data.equitytool.org SECRET_KEY='[your randomly-generated Django secret key]'
# Optional Python exception logging
dokku config:set data.equitytool.org RAVEN_DSN='[your Sentry DSN]'
    Connect the application to an instance of KoBoToolbox; see Development without Dokku for more information: 
    dokku config:set data.equitytool.org KPI_URL='https://kf.kobotoolbox.org'
dokku config:set data.equitytool.org KPI_API_KEY='[your kpi authorized application key]'
  5. Enable TLS (HTTPS): 
    dokku letsencrypt:enable data.equitytool.org
  6. Enable automatic renewal of TLS certificates: 
    dokku letsencrypt:cron-job --add
  7. Install Postgres and link it with the new application: 
    sudo dokku plugin:install https://github.com/dokku/dokku-postgres.git
dokku postgres:create reports
dokku postgres:link reports data.equitytool.org
  8. Add persistent storage for media uploads: 
    dokku storage:mount data.equitytool.org /var/lib/dokku/data/storage/data.equitytool.org:/app/media
  9. Follow the Dokku documentation to deploy the application code from your local machine to this new server using git push. This relies on the base image having already been built and pushed to Docker Hub by GitHub Actions. You must make a new release (or push a new tag) to trigger building of the base image. Reducing the size of this base image (currently over 4 GiB) would be a nice improvement.

Deploying an upgrade

  1. Make sure the master branch on GitHub has been updated with the code you want to deploy.
  2. Create a new release (or push a new tag) from the tip of master to trigger building of the base image by GitHub Actions.
  3. Wait for the build to complete. Verify that the base image for your new release is present on Docker Hub.
  4. Make sure your public SSH key has been added to authorized_keys on the production server.
  5. Use dokku ssh-keys:list to verify that your SSH key is added to dokku.
    • If the SSH key needs to be added, copy the .pub file on to the server and run dokku ssh-keys:add <keyname> <path/to/key>
  6. Create a new Git remote in your local copy of this repository, unless you've already set this up: 
    git remote add PRODUCTION [email protected]:data.equitytool.org
  7. Make sure the master branch in your local repository has been updated with the code you want to deploy.
  8. Deploy by pushing to the PRODUCTION remote: 
    git push PRODUCTION master
  9. Once you're satisfied with the deployment, you may want to prune unused Docker resources to save disk space: 
    docker system prune -a

Development without Dokku

This application requires a working instance of KoBoToolbox to run. See kobo-install for instructions on how to install such an instance.

  1. Go to https://[YOUR KPI DOMAIN]/admin/kpi/authorizedapplication/ (you will need to log in as a superuser);
  2. Click Add authorized application;
  3. Name your application and note the randomly-generated key (or enter your own);
  4. Click Save;
  5. Edit docker-compose.yml for this reports application:
    1. Set the KPI_API_KEY environment variable equal to the application key generated above;
    2. Set KPI_URL to https://[YOUR KPI DOMAIN]/;
      • If you are using a locally-hosted KoBoToolbox instance, you may need to configure extra_hosts as well.
    3. Set ALLOWED_HOSTS to match the hostname of your reports instance; see https://docs.djangoproject.com/en/3.2/ref/settings/#allowed-hosts.
  6. Execute docker compose pull;
  7. Execute docker build -t kobotoolbox/reports_base -f Dockerfile.base . (this is a slow process);
  8. Execute docker compose build;
    • WARNING: This builds a Docker image using the latest front-end code in your source tree, BUT the static files and Node dependencies built into the Docker image will be shadowed by the ./:/app volume in docker-compose.yml. You must run npm install and npm run build (or npm run dev) additionally, or else the application will run with stale code.
  9. Execute docker compose up -d postgres;
  10. Execute docker compose logs -f;
  11. Wait for the Postgres container to settle as indicated by the logs;
  12. Interrupt (CTRL+C) docker compose logs;
  13. Start the web application with docker compose up -d;
  14. Get a shell inside the application container by running docker compose exec koboreports bash;
  15. If desired, load some sample Forms into the database: 
    # Inside the application container
source activate koboreports
./manage.py loaddata dev/sample-forms.json
  16. To access the Django Admin interface, you'll need a superuser account. Create one now: 
    # Inside the application container
source activate koboreports
./manage.py createsuperuser
  17. Build the front-end files:
    1. On your host computer (not inside a Docker container), enter the jsapp directory within your source tree;
      • (Your source tree is mounted inside the Docker container by the ./:/app volume in docker-compose.yml)
    2. Use nvm or similar to run the same version of Node as specified in Dockerfile.base;
    3. Execute npm install;
    4. If you plan to do front-end development, execute npm run dev, which will watch your code for changes and reload as needed.
      • If you do this, you must visit the application at http://localhost:8080/. Accessing port 5000 will use stale front-end code.
    5. If you do not plan to touch front-end code, execute npm run build to rebuild the front-end static files one time only. You must do this at least once after switching branches or changing front-end dependencies, even if you do not edit any front-end code yourself.
  18. Access the application in a browser:
    • If you are not working on front-end code and have used only npm run build, access the application at http://localhost:5000/.
    • If you are modifying front-end code and executed npm run dev, you must access the application at http://localhost:8080/. Port 5000 will appear to work, but it will serve stale front-end code.