sql-runner 0.5.0

DEPT SQL runner

The DEPT SQL runner has three basic functionalities

• executing SQL code in a specific order

runner --execute {RUNNER_FILE_1}, {RUNNER_FILE_2} ..

• executing SQL code in a specific order, in staging mode (on test schema, tables and data)

runner --staging {RUNNER_FILE_1}, {RUNNER_FILE_2} ..

• quickly testing SQL code through temporary creation of views

runner --test {RUNNER_FILE_1}, {RUNNER_FILE_2} ..

• plotting of a dependency graph

runner --deps

An alias for the runner command is sqlrunner, for legacy purposes.

Using run_sql will run in interactive mode. run_sql /path/to/config.json

The supported databases are Redshift, Snowflake and Postgres.

Installation

SQL-Runner has the following optional dependencies that have to be mentioned when needed, during the installation process with pip:

• azuredwh - for work with Azure SQL Data Warehouse
• snowflake - for working with Snowflake DB
• redshift - for working with AWS Redshift
• bigquery - for working with Google BigQuery
• s3 - for enabling AWS S3 API access (for saving dependencies SVG graph)

Additionally for Azure DWH, it's required to install the Microsoft ODBC Driver. For Ubuntu 18.04 this is sufficient:

# In case any of these gest stuck, simply run `sudo su` once, to cache the password, then exit using Ctrl+D
curl https://packages.microsoft.com/keys/microsoft.asc | sudo apt-key add -
curl https://packages.microsoft.com/config/ubuntu/18.04/prod.list | sudo tee /etc/apt/sources.list.d/mssql-release.list > /dev/null
sudo apt-get update
sudo ACCEPT_EULA=Y apt-get install msodbcsql17
sudo apt-get install unixodbc-dev

Another dependency is graphviz:

sudo apt install graphviz

It is highly recommend it to install it in a virtual environment.

To create a virtual environment, run this:

sudo apt-get install python3-virtualenv
python3 -m virtualenv -p python3 venv

To install in a virtual environment, run this:

source venv/bin/activate
# Install with dependencies, ex. s3 and azuredwh
pip install git+https://github.com/leroi-marketing/sql-runner.git#egg=sql-runner[azuredwh]
# Or install from pypi
pip install sql-runner[azuredwh]

But if you really want to install it globally, run this:

sudo apt install python3-pip
# Install with dependencies, ex. s3 and azuredwh
sudo pip install git+https://github.com/leroi-marketing/sql-runner.git#egg=sql-runner[azuredwh]
# Or install from pypi
pip install sql-runner[azuredwh]

Configuration

Two configuration files are needed to use the sqlrunner.

• A config.json file that specifies all the necessary configuration variables. The default path is auth/config.json relative to the directory that this is run from.

{
    "sql_path": "{PATH}",
    "database_type": "[snowflake|redshift|postgres|bigquery|azuredwh]",
    "explicit_database": true if has to be present in every table reference (ex. snowflake)
    "auth": {
        // For Azure Synapse Analytics only
        "server": "url.of.azuredwh.server",
        // for BigQuery only
        "credentials_path": "/path/to/google-generated-credentials.json",

        // for Snowflake only
        "account": "{SNOWFLAKE_ACCOUNT}",

        // Azure Synapse Analytics DB, or Snowflake DB, or BigQuery Project ID
        "database": "{DATABASE}",

        // Postgresql or Redshift
        "dbname": "{POSTGRES_DATABASE} OR {REDSHIFT_DATABASE}",
        "host": "{POSTGRES_HOSTNAME} OR {REDSHIFT_HOSTNAME}",
        "port": "{POSTGRES_PORT} OR {REDSHIFT_PORT}"

        // Snowflake, postgres, redshift
        "user": "{USERNAME}",
        // Azure Synapse Analytics
        "username": "{USERNAME}",

        // All except Google BigQuery
        "password": "{PASSWORD}",
    },
    // configure staging environments as database suffix for all but the source data objects
    "staging": {
      "override": {
        "database": {
          "suffix": "_STAGING1"
        }
      },
      // python3 code that exposes `re` - regular expressions module, `database`, `schema`, `relation` being referenced
      "except": "not re.match('dwh', database.lower()) or re.search('^x', schema)"
    },
    // configure test schema creation locations as a schema prefix for all but the source data objects
    "test": {
      "override": {
        "schema": {
          "prefix": "zz_"
        }
      },
      // python3 code that exposes `re` - regular expressions module, `database`, `schema`, `relation` being referenced
      "except": "not re.match('dwh', database.lower()) or re.search('^x', schema)"
    },
    // Add a dependency cache file, to speed up run initialization
    "deps_cache": {
      "type": "filesystem",
      "location": "/path/to/local/cache/dependencies.csv"
    },
    "deps_schema": "{DEPENDENCY_SCHEMA_NAME}",
    "exclude_dependencies": [
        "EXCLUDED_SCHEMA_1",
        "EXCLUDED_SCHEMA_2"
    ],
    "graphviz_path": "{GRAPHVIZ_PATH_FOR_WINDOWS}"
}

Alternatively, a path to a Python script that gets included into sqlrunner can also be supplied. The script has to have a class Config with static value members or @property members for every JSON main property. Short, insufficient example:

class Config:
    sql_path = "sql"
    database_type = "snowflake"
    explicit_database = True
    test = {
        "override": {
            "schema": {
                "prefix": "zz_"
            }
        },
        "except": "re.search('^x', schema)"
    }

    @property
    def auth(self):
        # Retrieve credentials from somewhere
        return {
            "user": "DEPT",
            "password": "123456",
            "database": "DWH",
            "account": "db"
        }


if __name__ == '__main__':
    import json
    config = {}
    config_obj = Config()
    for key in dir(config_obj):
        if not key.startswith('__'):
            config[key] = getattr(config_obj, key)
    print(json.dumps(config, indent=4))

This feature allows one to store sensitive credentials in an encrypted state

• One or more csv files specifying the name of the the tables and views and their respective schemas.

{SCHEMA_1};{SQL_FILENAME_1};e
{SCHEMA_1};{SQL_FILENAME_2};e
{SCHEMA_1};{SQL_FILENAME_3};e
{SCHEMA_2};{SQL_FILENAME_4};e
{SCHEMA_3};{SQL_FILENAME_5};e
..

Per schema one directory is expected. The name of the SQL files should correspond to thename of the respective table or view. The last columns specifies the desired action.

e: execute the query
t: create table
v: create view
m: materialize view
check: run assertions on query result

Development

To set up dependencies locally for development:

# Install virtualenv (if your default python is python2, specify also `-p python3`)
python3 -m virtualenv -p python3 venv
source venv/bin/activate
pip install -e .[azuredwh] # and other optional dependencies

# Run local (non-build) version:
python debug.py [arg1 arg2 ...]

Functional comments

Queries can have functional comments on the top. These comments can either specify data distribution for Azure Synapse Analytics or RedShift, or can contain assertions for check queries.

Check queries

Adding a functional comment at the top of the sql file, in the form of:

/*
assert_row_count 0
*/
SELECT 1 FROM my_schema.my_table WHERE revenue < 0;

Gives you the option to synthetically fail a step if the returned rows don't correspond to the expectation. There are currently 2 tests supported but they can easily be extended:

• assert_row_count <x> - fails if the number of rows returned by the statement is different from x
• assert_almost_equal <tolerance value> - fails if the 2 rows returned with single columns have values that differ from each other by more than tolerance value

To add more tests, check out sql_runner/tests.py

Override dependencies

Sometimes you want to just update a table, not re-create it. This calls for an execute type query, and the UPDATE itself isn't well parsed by the dependency detector. For that, and other cases where dependency detection doesn't work to your service, you can help it with these functional comments.

Anywhere in the SQL statement, add a comment that has valid JSON. The following JSON keys are currently supported:

• "node_id": ["my_schema", "my_table"] - overrides the name from the query list CSV and from the file name. This lets you have multiple steps that work on the same table
• "override_dependencies": [["my_schema", "mytable1"], ["my_schema", "mytable2"]] - tells the dependency parser to completely ignore the query when detecting dependencies, and to take only these
• "ignore_dependencies": [["my_schema", "mytable1"], ["my_schema", "mytable2"]] - tells the dependency parser to ignore a list of dependencies from the ones detected in the query.
• "additional_dependencies": [["my_schema", "mytable1"], ["my_schema", "mytable2"]] - tells the dependency parser to also include a list of explicit dependencies on top of the ones already detected.

Preprocess names in e statements

"execute" e statements in legacy versions were not processed at all to substitute names. With the addition of the "preprocess_names": true value, sources and destinations will be updated accordingly (staging prefix, suffix, etc).

This needs better documentation, but for now you can check the source code for the DB-specific Query classes in sql_runner/db.