Skip to content
2018 repo for the transportation api backend
Branch: staging
Clone or download
Latest commit fd17171 May 21, 2018

README.md

Hack Oregon Transportation Systems Backend Repository

This repo represents the work of the Transportation Systems project of Hack Oregon. We are a volunteer project for Open Data.

This repo is intended to be run in a docker environment.

A note about line endings

As you probably know, text files on Linux (where we deploy and where our containers run) have by convention lines ending with LF. However, Windows, where some of us test and develop, uses the convention that lines in a text file end with a CR and a LF.

Git knows what we're using, and it tries to accomodate us by checking out working files with the line endings our platform uses. See this explainer from GitHub on how line endings work with Git: https://help.github.com/articles/dealing-with-line-endings/.

Now, throw in Docker and Docker Compose. Some Dockerfiles call for copying files from your host into the image during the build. If those files came from a Git repo, the default is that their line endings are your host native convention. But inside the image, which is a Linux filesystem, some files must use the Linux line ending convention or they won't work.

So far, we know that Bash and sh scripts will crash if they have Windows line endings, often with mysterious error messages. Also, Python scripts like Django's manage.py, when executed as commands, will crash mysteriously if they have Windows line endings.

One final note: vim has a command :setfileformat. Either :setfileformat unix if you want the Unix / Linux convention or :setfileformat dos if you want the DOS / Windows convention. Then save the file and edit your .gitattributes file to declare that the file must have that convention on checkout.

Restarting Docker for Windows sometimes necessary

Sometimes, Docker for Windows loses contact with some critical resource and throws ugly messages like this:

ERROR: for transportationsystembackend_db_1  Cannot start service db: driver failed programming external connectivity on endpoint transportationsystembackend_db_1 (f944aeb0244747359af77373b4949561c6e6e1d8ee48fb0bfc8aba98aa32877e): Error starting userland proxy: mkdir /port/tcp:0.0.0.0:5439:tcp:172.18.0.2:5432: input/output error

ERROR: for db  Cannot start service db: driver failed programming external connectivity on endpoint transportationsystembackend_db_1 (f944aeb0244747359af77373b4949561c6e6e1d8ee48fb0bfc8aba98aa32877e): Error starting userland proxy: mkdir /port/tcp:0.0.0.0:5439:tcp:172.18.0.2:5432: input/output error
Encountered errors while bringing up the project.

If this happens, you will need to restart Docker. Open the Settings dialog and go to Reset. Select the Restart option (the top one). Wait till the green Docker is running light shows up and then go back to your terminal. Everything should then work. This is a known Docker for Windows bug, not something you did wrong.

Getting Started

In order to run this you will want to:

  1. Clone this Repository

  2. cd into it

  3. The environment variables that Docker uses and inserts into the images it builds are taken from a file in the root of this repository called .env. Because it contains sensitive information like passwords, it is not checked into version control - you have to create it as follows:

    • Copy env.sample to .env: cp env.sample .env
    • Edit .env and change at least POSTGRES_PASSWORD and DJANGO_SECRET_KEY. You should not need to change any of the others during test and development.
  4. Download the database .backup files from Google Drive and place them in ./Backups before doing the Docker build. The build will copy them onto the image and the first "run" in a container will restore them. See Automatic database restores for the details on the restore mechanism.

  5. The .env file and the .backup files have been added to the .gitignore file. Provided you do not rename them or change locations they will not be committed to the repo and this project will build and run.

  6. Confirm you have executable perms on all the scripts in the ./bin folder: $ chmod +x ./bin/*.sh Feel free to read each one and assign perms individually, cause it is your computer 😜 and security is a real thing.

  7. Run the build.sh script to build the project. Since you are going to be running it on the local machine you will want to run: ./bin/build.sh -l - This command is doing a docker-compose build in the background. It is downloading the images needed for the project to your local machine.

  8. Once this completes you will now want to start up the project. We will use the start.sh script for this, again using the -l flag to run locally: ./bin/start.sh -l The first time you run this you will see the database restores. You can ignore the error messages. You will also see the api container start up.

  9. Once the first startup completes kill the container using cmd c/ctrl c depending on your os.

  10. Restart the container using the same start command: ./bin/start.sh -l and both the db and the api will start up.

  11. Open your browser and you will be able to access the Django Rest Framework browserable front end at http://localhost:8000/api, the Swagger API schema at http://localhost:8000/schema, and the Django admin login at http://localhost:8000/admin.

  12. To Run Tests: run the ./bin/build.sh -l followed by the ./bin/test.sh -l command.

  13. Note that the api container will write some files into your Git repository. They're in .gitignore, so they won't be checked into version control.

Run in Staging Environment

While developing the API, using the built in dev server is useful as it allows for live reloading, and debug messages. When running in a production environment, this is a security risk, and not efficient. As such a staging/production environment has been created using the following technologies:

  • Gunicorn - A "green" HTTP server
  • Gevent - Asynchronous workers
  • Pyscopgreen - A "green" version of the psycop database connector
  • django_db_geventpool - DB pool using gevent for PostgreSQL DB.
  • WhiteNoise - allows for hosting of static files by gunicorn in a prod environment vs. integrating a webserver

Instructions:

  1. copy the /bin/env.staging.sample file to create a .env.staging file in same directory:
$ cp ./bin/env.staging.sample ./bin/.env.staging
  1. open the ./bin/.env in your text editor and complete the environmental variables.

  2. Download and save the sql file if you have not already.

  3. Run the build.sh script to build the project for the staging environment: $ ./bin/build.sh -s

  4. Start the project using the staging flag: $ ./bin/start.sh -s

  5. Open your browser and you should be able to access the Django Restframework browserable front end at: http://localhost:8000/api and Swagger at http://localhost:8000/schema

  6. Try going to an nonexistent page and you should see a generic 404 Not found page instead of the Django debug screen.

What was configured:

So what is changed from the default Django setup for the staging environment. This already has been done, being included for informational purposes

  • Add gunicorn, gevent, and whitenoise, django-db-geventpool, to requirements.txt
  • Set the debug variable to false in the .env (In future maybe setup a separate env variable for environment?)
  • make any other changes necessary to config vars, ie: database settings
  • create a staging entrypoint/prod entrypoint file that runs the gunicorn start command instead of the ./manage.py runserver. Here is an example:
gunicorn crash_data_api.wsgi -c gunicorn_config.py
  • create the gunicorn_config.py file to hold gunicorn config, including using gevent worker_class. Currently we are patching psycopg2 and django with gevent/psycogreen in the post_fork worker. Also using 4 workers:
try:
    # fail 'successfully' if either of these modules aren't installed
    from gevent import monkey
    from psycogreen.gevent import patch_psycopg


    # setting this inside the 'try' ensures that we only
    # activate the gevent worker pool if we have gevent installed
    worker_class = 'gevent'
    workers = 4
    # this ensures forked processes are patched with gevent/gevent-psycopg2
    def do_post_fork(server, worker):
        monkey.patch_all()
        patch_psycopg()

        # you should see this text in your gunicorn logs if it was successful
        worker.log.info("Made Psycopg2 Green")

    post_fork = do_post_fork
except ImportError:
    pass
  • Change the settings.py file to use django_db_geventpool when in production mode. You will add this after the current database settings.:
if os.environ.get('DEBUG') == "False":

    DATABASES = {
        'default': {
            'ENGINE': 'django_db_geventpool.backends.postgis',
            'PASSWORD': os.environ.get('POSTGRES_PASSWORD'),
            'NAME': os.environ.get('POSTGRES_NAME'),
            'USER': os.environ.get('POSTGRES_USER'),
            'HOST': os.environ.get('POSTGRES_HOST'),
            'PORT': os.environ.get('POSTGRES_PORT'),
            'CONN_MAX_AGE': 0,
            'OPTIONS': {
                'MAX_CONNS': 20
            }
        }
    }
  • create a staging/production docker_compose file, to use the correct variables from the .env file, entrypoint, and any other changes needed
  • Make changes to settings.py to check the debug variable and use :

Change DEBUG line:

DEBUG = os.environ.get('DEBUG') == "True" - handles os variables being treated as strings

ADD to MIDDLEWARE right after SECURITY:

'whitenoise.middleware.WhiteNoiseMiddleware',

ADD these just before the STATIC_URL so staticfiles are handled correctly and are compressed:


STATICFILES_STORAGE = 'whitenoise.storage.CompressedManifestStaticFilesStorage'

STATIC_ROOT = os.path.join(BASE_DIR, 'staticfiles')

Contributing

To develop on the repo,

  1. Create an issue for tracking and communication

  2. You will clone repo and then create a feature branch.

  3. After branching confirm you can follow above get started steps.

  4. Develop you feature

  5. Update documentation and env sample file as necessary.

  6. Commit Changes.

  7. Merge current Staging branch into feature branch to resolve any merge conflicts.

  8. Push local feature branch to Hack Oregon repo. Any PR requests from forks will be rejected.

  9. Create a Pull Request to staging branch. No PRs will be accepted to Master unless from staging and by approved reviewers

  10. PR should be reviewed by authorized reviewer, another team member if possible, and pass any automated testing requirements.

  11. Any outstanding merge conflicts resolved

  12. Authorized reviewer will commit to staging.

  13. Process for staging to master will be defined.

API

The primary function of this API is to act as a read-only wrapper around ODOT's Crash data and expose the underlying data to the web via HTTP Requests. The secondary function is eventually expose helper functions that could simplify data pre-processing via in-built helper functions. This API aims to be RESTful.

Note on Unmanaged models

The models in this project are unmanaged. Given that a) the API sits upon a legacy database and b) the API is intended to be read-only, the decision was made to decouple Django from database management and isolate that solely to the underlying PostGres shell environment. This is to prevent creation and deletions of the underlying data tables primarily during development. Malicious editing (outside of the dev environment) is less of a concern since that can be handled by a secure permissions for users making API calls.

Note on Permissions

All users can browse the API. Read-only access is the default permission for unauthenticated users.

Note on Testing

Testing an unmanaged model requires a few modifications to the test runner. Since migrations don't create any tables, they create a blank test database which results in no test data being found. The fix is outlined in the following post - https://dev.to/patrnk/testing-against-unmanaged-models-in-django

Runnning a test requires you have 'django-test-without-migrations' as part of your requirements. The only other point to remember is that tests need to be run with ./manage.py test --no-migrations flag to prevent Django from trying to run migrations on your test db.

Endpoints

Crashes Table

TBD

Participants Table

TBD

Vehicles Table

TBD

Filtering

Three types of filters are currently supported -

1. Search Filters

Simple text search can be performed on the following fields:

Crash Table
'crash_id','crash_hr_short_desc','urb_area_short_nm','fc_short_desc','hwy_compnt_short_desc','mlge_typ_short_desc', 'specl_jrsdct_short_desc','jrsdct_grp_long_desc','st_full_nm','isect_st_full_nm','rd_char_short_desc', 'isect_typ_short_desc','crash_typ_short_desc','collis_typ_short_desc','rd_cntl_med_desc','wthr_cond_short_desc','rd_surf_short_desc','lgt_cond_short_desc','traf_cntl_device_short_desc','invstg_agy_short_desc','crash_cause_1_short_desc','crash_cause_2_short_desc','crash_cause_3_short_desc','pop_rng_med_desc','rd_cntl_med_desc'
Participants Table

TBD

Vehicles Table

TBD

Usage:

To look for all fields listed above that match (not exact) the string "DIS-RAG" -

http://localhost:8000/api/crashes/?search=DIS--RAG

2. Field Filters

The API also supports explicit filter fields as part of URL query strings. The following fields are currently supported -

'ser_no','cnty_id','alchl_invlv_flg','crash_day_no','crash_mo_no','crash_yr_no','crash_hr_no','schl_zone_ind','wrk_zone_ind','alchl_invlv_flg','drug_invlv_flg','crash_speed_invlv_flg','crash_hit_run_flg'

Note:

  1. Filters need to be an exact match.
  2. URLs need to be encoded. All reserved characters characters should be escaped. For example, a query on the field public_location_description with the value SW 6th & Salmon, then the query should look like this - http://localhost:8000/api/passenger-census/?public_location_description=SW%206th%20%26%20Salmon. Here both spaces (%20) and ampersands (%26) have been escaped.
Usage:

If filtering just "00173" and "00174" for the field 'ser_no' -

http://localhost:8000/api/crashes/?ser_no=00173&ser_no=00174

3. Ordering Filters

Results can be sorted against any field or combinations of fields.

Usage:

To show results in ascending order of the field 'ser_no':

http://localhost:8000/api/crashes/?ordering=ser_no

In descending order:

http://localhost:8000/api/crashes/?ordering=-ser_no

multiple fields:

http://localhost:8000/api/crashes/?ordering=-ser_no,rd_cntl_med_desc

Versions

The API supports Accept Header Versioning. Version numbers in API requests are optional and if no version is specified the request header latest version is returned by default. Specify versions as numbers, as shown in header example below -

GET /api/crashes HTTP/1.1
Host: example.com:8000
Accept: application/json; version=1.0

Latest version: 1.0 (as of 02/19/2018)

License

We follow the MIT License: https://github.com/hackoregon/transportation-system-backend/blob/staging/LICENSE

Contributors

You can’t perform that action at this time.