graphql-engine/server/CONTRIBUTING.md

Contributing

This guide explains how to set up the graphql-engine server for development on your own machine and how to contribute.

Pre-requisites

• GHC 8.10.2 and cabal-install
  • There are various ways these can be installed, but ghcup is a good choice if you’re not sure.
• There are few system packages required like libpq-dev, libssl-dev, etc. The best place to get the entire list is from the packager Dockerfile

For building console and running test suite:

• Node.js (v12+, it is recommended that you use node with version v12.x.x A.K.A erbium or version 14.x.x A.K.A Fermium)
• npm >= 5.7
• python >= 3.5 with pip3 and virtualenv

Additionally, you will need a way to run a Postgres database server. The dev.sh script (described below) can set up a Postgres instance for you via Docker, but if you want to run it yourself, you’ll need:

• PostgreSQL >= 9.5
• postgis

Upgrading npm

If your npm is too old (>= 5.7 required):

$ npm install -g npm@latest   # sudo may be required

or update your nodejs.

Development workflow

You should fork the repo on github and then git clone https://github.com/<your-username>/graphql-engine. After making your changes

Compile

...console assets:

$ cd console
$ npm ci
$ npm run server-build
$ cd ..

...and the server:

$ ln -s cabal.project.dev cabal.project.local
$ cabal new-update
$ cabal new-build graphql-engine

To set up the project configuration to coincide with the testing scripts below, thus avoiding recompilation when testing locally, rather use cabal.project.dev-sh.local instead of cabal.project.dev:

$ ln -s cabal.project.dev-sh.local cabal.project.local

IDE Support

You may want to use hls/ghcide if your editor has LSP support. A sample configuration has been provided which can be used as follows:

ln -s sample.hie.yaml hie.yaml

If you have to customise any of the options for ghcide/hls, you should instead copy the sample file and make necessary changes in hie.yaml file. Note that hie.yaml is gitignored so the changes will be specific to your machine.

cp sample.hie.yaml hie.yaml

Run and test via dev.sh

The dev.sh script in the top-level scripts/ directory is a turnkey solution to build, run, and test graphql-engine using a Docker container to run a Postgres database. Docker is necessary to use dev.sh.

To use dev.sh, first launch a new postgres container with:

$ scripts/dev.sh postgres

Then in a new terminal launch graphql-engine in dev mode with:

$ scripts/dev.sh graphql-engine

The dev.sh will print some helpful information and logs from both services will be printed to screen.

You can run the test suite with:

$ scripts/dev.sh test

This should run in isolation. The output format is described in the pytest documentation. Errors and failures are indicated by Fs and Es.

Optionally, launch a new container for alternative (MSSQL) backend with:

$ scripts/dev.sh mssql

Tests can be run against a specific backend (defaulting to Postgres) with the backend flag, for example:

$ scripts/dev.sh test --integration -k TestGraphQLQueryBasicCommon --backend (bigquery|citus|mssql|postgres)

Run and test manually

If you want, you can also run the server and test suite manually against an instance of your choosing.

Run

The following command can be used to build and launch a local graphql-engine instance:

cabal new-run -- exe:graphql-engine \
  --database-url='postgres://<user>:<password>@<host>:<port>/<dbname>' \
  serve --enable-console --console-assets-dir=console/static/dist

This will launch a server on port 8080, and it will serve the console assets if they were built with npm run server-build as mentioned above.

Test

graphql-engine has several test suites, among them:

1. A small set of unit tests and integration tests written in Haskell, in server/src-test.

2. An extensive set of end-to-end tests written in Python, in server/tests-py.

Both sets of tests require a running Postgres database.

Running the Haskell test suite

cabal new-run -- test:graphql-engine-tests unit
HASURA_GRAPHQL_DATABASE_URL='postgres://<user>:<password>@<host>:<port>/<dbname>' \
    cabal new-run -- test:graphql-engine-tests postgres

Running the Python test suite

1. To run the Python tests, you’ll need to install the necessary Python dependencies first. It is recommended that you do this in a self-contained Python venv, which is supported by Python 3.3+ out of the box. To create one, run:

   python3 -m venv .python-venv
   

   (The second argument names a directory where the venv sandbox will be created; it can be anything you like, but .python-venv is .gitignored.)

   With the venv created, you can enter into it in your current shell session by running:

   source .python-venv/bin/activate
   

   (Source .python-venv/bin/activate.fish instead if you are using fish as your shell.)

2. Install the necessary Python dependencies into the sandbox:

   pip3 install -r tests-py/requirements.txt
   

3. Install the dependencies for the Node server used by the remote schema tests:

   (cd tests-py/remote_schemas/nodejs && npm ci)
   

4. Start an instance of graphql-engine for the test suite to use:

   env EVENT_WEBHOOK_HEADER=MyEnvValue \
       WEBHOOK_FROM_ENV=http://localhost:5592/ \
       SCHEDULED_TRIGGERS_WEBHOOK_DOMAIN=http://127.0.0.1:5594 \
     cabal new-run -- exe:graphql-engine \
       --database-url='postgres://<user>:<password>@<host>:<port>/<dbname>' \
       serve --stringify-numeric-types
   

   Optionally, replace the --database-url parameter with --metadata-database-url to enable testing against multiple sources.

   The environment variables are needed for a couple tests, and the --stringify-numeric-types option is used to avoid the need to do floating-point comparisons.

5. Optionally, add alternative sources to test against:

   If you enabled testing against multiple sources with in the last step, you can add those sources as follows:

   # Add a Postgres source
   curl "$METADATA_URL" \
   --data-raw '{"type":"pg_add_source","args":{"name":"default","configuration":{"connection_info":{"database_url":"'"$POSTGRES_DB_URL"'","pool_settings":{}}}}}'
   
   # Add a SQL Server source
   curl "$METADATA_URL" \
   --data-raw '{"type":"mssql_add_source","args":{"name":"mssql","configuration":{"connection_info":{"connection_string":"'"$MSSQL_DB_URL"'","pool_settings":{}}}}}'
   
   # Optionally verify sources have been added
   curl "$METADATA_URL" --data-raw '{"type":"export_metadata","args":{}}'
   

6. With the server running, run the test suite:

   cd tests-py
   pytest --hge-urls http://localhost:8080 \
          --pg-urls 'postgres://<user>:<password>@<host>:<port>/<dbname>'
   

This will run all the tests, which can take a couple minutes (especially since some of the tests are slow). You can configure pytest to run only a subset of the tests; see the pytest documentation for more details.

Some other useful points of note:

• It is recommended to use a separate Postgres database for testing, since the tests will drop and recreate the hdb_catalog schema, and they may fail if certain tables already exist. (It’s also useful to be able to just drop and recreate the entire test database if it somehow gets into a bad state.)

• You can pass the -v or -vv options to pytest to enable more verbose output while running the tests and in test failures. You can also pass the -l option to display the current values of Python local variables in test failures.

• Tests can be run against a specific backend (defaulting to Postgres) with the backend flag, for example:

    pytest --hge-urls http://localhost:8080 \
           --pg-urls 'postgres://<user>:<password>@<host>:<port>/<dbname>'
           --backend mssql -k TestGraphQLQueryBasicCommon
  

Running the Python test suite on BigQuery

Running integration tests against a BigQuery data source is a little more involved due to the necessary service account requirements:

HASURA_BIGQUERY_PROJECT_ID=# the project ID of the service account
HASURA_BIGQUERY_SERVICE_ACCOUNT_EMAIL=# eg. "<<SERVICE_ACCOUNT_NAME>>@<<PROJECT_NAME>>.iam.gserviceaccount.com"
HASURA_BIGQUERY_SERVICE_ACCOUNT_FILE=# the filepath to the downloaded service account key

Before running the test suite either manually or via dev.sh:

1. Ensure you have access to a Google Cloud Console service account. Store the project ID and account email in HASURA_BIGQUERY_PROJECT_ID and (optional) HASURA_BIGQUERY_SERVICE_ACCOUNT_EMAIL variables.
2. Create and download a new service account key. Store the filepath in a HASURA_BIGQUERY_SERVICE_ACCOUNT_FILE variable.
3. Login and activate the service account, if it is not already activated.
4. Verify the service account is accessible via the BigQuery API:
   1. Run source scripts/verify-bigquery-creds.sh $HASURA_BIGQUERY_PROJECT_ID $HASURA_BIGQUERY_SERVICE_ACCOUNT_FILE $HASURA_BIGQUERY_SERVICE_ACCOUNT_EMAIL. If the query succeeds, the service account is setup correctly to run tests against BigQuery locally.
5. Finally, run the BigQuery test suite with HASURA_BIGQUERY_SERVICE_ACCOUNT_FILE and HASURA_BIGQUERY_PROJECT_ID environment variables set. For example:

scripts/dev.sh test --integration --backend bigquery -k TestGraphQLQueryBasicBigquery

Guide on writing python tests

1. Check whether the test you intend to write already exists in the test suite, so that there will be no duplicate tests or the existing test will just need to be modified.

2. All the tests use setup and teardown, the setup step is used to initialize the graphql-engine and the database in a certain state after which the tests should be run. After the tests are run, the state needs to be cleared, which should be done in the teardown step. The setup and teardown is localised for every python test class.

   See TestCreateAndDelete in test_events.py for reference.

3. The setup and teardown can be configured to run before and after every test in a test class or run before and after running all the tests in a class. Depending on the use case, there are different fixtures like per_class_tests_db_state,per_method_tests_db_state defined in the conftest.py file.

4. Sometimes, it's required to run the graphql-engine with in a different configuration only for a particular set of tests. In this case, these tests should be run only when the graphql-engine is run with the said configuration and should be skipped in other graphql-engine configurations. This can be done by accepting a new command-line flag from the pytest command and depending on the value or presence of the flag, the tests should be run accordingly. After adding this kind of a test, a new section needs to be added in the test-server.sh. This new section's name should also be added in the server-test-names.txt file, otherwise the test will not be run in the CI.

   For example,

   The tests in the test_remote_schema_permissions.py are only to be run when the remote schema permissions are enabled in the graphql-engine and when it's not set, these tests should be skipped. Now, to run these tests we parse a command line option from pytest called (--enable-remote-schema-permissions) and the presence of this flag means that we need to run these tests. When the tests are run with this command line option, it's assumed that the server has enabled remote schema permissions.

Adding test support for a new backend

The current workflow for supporting a new backend in integration tests is as follows:

1. Add functions to launch and cleanup a server for the new backend. Example.
2. Connect to the database you've just launched. Example.
3. Add setup and teardown files:
   1. setup_<backend>: for v1/query or metadata queries such as <backend>_track_table. Example.
   2. schema_setup_<backend>: for v2/query queries such as <backend>_run_sql. Example.
   3. teardown_<backend> and cleardb_<backend>
   4. important: filename suffixes should be the same as the value that’s being passed to —backend; that's how the files are looked up.
4. Specify a backend parameter for the per_backend_tests fixture, parameterised by backend. Example.

Note: When teardown is not disabled (via skip_teardown, in which case, this phase is skipped entirely), teardown.yaml always runs before schema_teardown.yaml, even if the tests fail. See setup_and_teardown in server/tests-py/conftest.py for the full source code/logic.

This means, for example, that if teardown.yaml untracks a table, and schema_teardown.yaml runs raw SQL to drop the table, both would succeed (assuming the table is tracked/exists).

Test suite naming convention The current convention is to indicate the backend(s) tests can be run against in the class name. For example: * TestGraphQLQueryBasicMySQL for tests that can only be run on MySQL * TestGraphQLQueryBasicCommon for tests that can be run against more than one backend * if a test class doesn't have a suffix specifying the backend, nor does its name end in Common, then it is likely a test written pre-v2.0 that can only be run on Postgres

This naming convention enables easier test filtering with pytest command line flags.

The backend-specific and common test suites are disjoint; for example, run pytest --integration -k "Common or MySQL" --backend mysql to run all MySQL tests.

Building with profiling

To build with profiling support, you need to both enable profiling via cabal and set the profiling flag. E.g.

cabal build exe:graphql-engine -f profiling --enable-profiling

Create Pull Request

• Make sure your commit messages meet the guidelines.
• If you changed the versions of any dependencies, run cabal new-freeze to update the freeze file.
• Create a pull request from your forked repo to the main repo.
• Every pull request will automatically build and run the tests.

Code conventions

The following conventions help us maintain a uniform style for all committers: make sure your contributions are in line with them.

We enforce these by means of CI hooks which will fail the build if any of these are not met.

• No compiler warnings: Make sure your code builds with no warnings (adding -Werror to ghc-options in your cabal.project is a good way of checking this.)
• No lint failures: Use hlint with our custom config to validate your code, using hlint --hint=server/.hlint.yaml.
• Consistent formatting: Use ormolu to format your code. ormolu -ei '*.hs' will format all files with a .hs extension in the current directory.
• Consistent style: Consider the style guide when writing new code.