Airflow ClickHouse Plugin

Provides ClickHouseOperator, ClickHouseHook and ClickHouseSqlSensor for Apache Airflow based on mymarilyn/clickhouse-driver.

Features

SQL queries are templated.
Can run multiple SQL queries per single ClickHouseOperator.
Result of the last query of ClickHouseOperator instance is pushed to XCom.
Executed queries are logged in a pretty form.
Uses effective native ClickHouse TCP protocol thanks to clickhouse-driver. Does not support HTTP protocol.
Supports extra ClickHouse connection parameters such as various timeouts, compression, secure, etc through Airflow Connection.extra property.

Installation and dependencies

pip install -U airflow-clickhouse-plugin

Requires apache-airflow and clickhouse-driver. Primarily supports Airflow 1.10.6 since it is the latest version supported by Google Cloud Composer: later versions are expected to work properly but may be not fully tested.

Usage

See examples below.

ClickHouseOperator Reference

To import ClickHouseOperator use: from airflow.operators.clickhouse_operator import ClickHouseOperator

Supported kwargs:

sql: templated query (if argument is a single str) or queries (if iterable of str's).
clickhouse_conn_id: connection id. Connection schema is described below.
parameters: passed to clickhouse-driver execute method.
- If multiple queries are provided via sql then the parameters are passed to all of them.
- Parameters are not templated.
database: if present, overrides database defined by connection.
Other kwargs (including the required task_id) are inherited from Airflow BaseOperator.

The result of the last query is pushed to XCom.

See example below.

ClickHouseHook Reference

To import ClickHouseHook use: from airflow.hooks.clickhouse_hook import ClickHouseHook

Supported kwargs of constructor (__init__ method):

clickhouse_conn_id: connection id. Connection schema is described below.
database: if present, overrides database defined by connection.

Supports all of the methods of the Airflow BaseHook including:

get_records(sql: str, parameters: dict=None): returns result of the query as a list of tuples. Materializes all the records in memory.
get_first(sql: str, parameters: dict=None): returns the first row of the result. Does not load the whole dataset into memory because of using execute_iter. If the dataset is empty then returns None following fetchone semantics.
run(sql, parameters): runs a single query (specified argument of type str) or multiple queries (if iterable of str). parameters can have any form supported by execute method of clickhouse-driver.
- If single query is run then returns its result. If multiple queries are run then returns the result of the last of them.
- If multiple queries are given then parameters are passed to all of them.
- Materializes all the records in memory (uses simple execute but not execute_iter).
  - To achieve results streaming by execute_iter use it directly via hook.get_conn().execute_iter(…) (see execute_iter reference).
- Every run call uses a new connection which is closed when finished.
get_conn(): returns the underlying clickhouse_driver.Client instance.

See example below.

ClickHouseSqlSensor Reference

Sensor fully inherits from Airflow SQLSensor and therefore fully implements its interface using ClickHouseHook to fetch the SQL execution result and supports templating of sql argument.

See example below.

ClickHouse Connection schema

clickhouse_driver.Client is initiated with attributes stored in Airflow Connection attributes. The mapping of the attributes is listed below:

Airflow Connection attribute	`Client.__init__` argument
`host`	`host`
`port`	`port`
`schema`	`database`
`login`	`user`
`password`	`password`

If you pass database argument to ClickHouseOperator or ClickHouseHook explicitly then it is passed to the Client instead of the schema attribute of the Airflow connection.

Extra arguments

You may also pass additional arguments, such as timeouts, compression, secure, etc through Connection.extra attribute. The attribute should contain a JSON object which will be deserialized and all of its properties will be passed as-is to the Client.

For example, if Airflow connection contains extra={"secure":true} then the Client.__init__ will receive secure=True keyword argument in addition to other non-empty connection attributes.

Default values

If the Airflow connection attribute is not set then it is not passed to the Client at all. In that case the default value of the corresponding clickhouse_driver.Connection argument is used (e.g. user defaults to 'default').

This means that Airflow ClickHouse Plugin does not itself define any default values for the ClickHouse connection. You may fully rely on default values of the clickhouse-driver version you use. The only exception is host: if the attribute of Airflow connection is not set then 'localhost' is used.

Default connection

By default the plugin uses connection_id='clickhouse_default'.

Examples

ClickHouseOperator Example

from airflow import DAG
from airflow.operators.clickhouse_operator import ClickHouseOperator
from airflow.operators.python_operator import PythonOperator
from airflow.utils.dates import days_ago

with DAG(
        dag_id='update_income_aggregate',
        start_date=days_ago(2),
) as dag:
    ClickHouseOperator(
        task_id='update_income_aggregate',
        database='default',
        sql=(
            '''
                INSERT INTO aggregate
                SELECT eventDt, sum(price * qty) AS income FROM sales
                WHERE eventDt = '{{ ds }}' GROUP BY eventDt
            ''', '''
                OPTIMIZE TABLE aggregate ON CLUSTER {{ var.value.cluster_name }}
                PARTITION toDate('{{ execution_date.format('%Y-%m-01') }}')
            ''', '''
                SELECT sum(income) FROM aggregate
                WHERE eventDt BETWEEN
                    '{{ execution_date.start_of('month').to_date_string() }}'
                    AND '{{ execution_date.end_of('month').to_date_string() }}'
            ''',
            # result of the last query is pushed to XCom
        ),
        clickhouse_conn_id='clickhouse_test',
    ) >> PythonOperator(
        task_id='print_month_income',
        provide_context=True,
        python_callable=lambda task_instance, **_:
            # pulling XCom value and printing it
            print(task_instance.xcom_pull(task_ids='update_income_aggregate')),
    )

ClickHouseHook Example

from airflow import DAG
from airflow.hooks.clickhouse_hook import ClickHouseHook
from airflow.hooks.mysql_hook import MySqlHook
from airflow.operators.python_operator import PythonOperator
from airflow.utils.dates import days_ago


def mysql_to_clickhouse():
    mysql_hook = MySqlHook()
    ch_hook = ClickHouseHook()
    records = mysql_hook.get_records('SELECT * FROM some_mysql_table')
    ch_hook.run('INSERT INTO some_ch_table VALUES', records)


with DAG(
        dag_id='mysql_to_clickhouse',
        start_date=days_ago(2),
) as dag:
    dag >> PythonOperator(
        task_id='mysql_to_clickhouse',
        python_callable=mysql_to_clickhouse,
    )

Important note: don't try to insert values using ch_hook.run('INSERT INTO some_ch_table VALUES (1)') literal form. clickhouse-driver requires values for INSERT query to be provided via parameters due to specifics of the native ClickHouse protocol.

ClickHouseSqlSensor Example

from airflow import DAG
from airflow.sensors.clickhouse_sql_sensor import ClickHouseSqlSensor
from airflow.operators.clickhouse_operator import ClickHouseOperator
from airflow.utils.dates import days_ago


with DAG(
        dag_id='listen_warnings',
        start_date=days_ago(2),
) as dag:
    dag >> ClickHouseSqlSensor(
        task_id='poke_events_count',
        database='monitor',
        sql="SELECT count() FROM warnings WHERE eventDate = '{{ ds }}'",
        success=lambda cnt: cnt > 10000,
    ) >> ClickHouseOperator(
        task_id='create_alert',
        database='alerts',
        sql='''
            INSERT INTO events SELECT eventDate, count()
            FROM monitor.warnings WHERE eventDate = '{{ ds }}'
        ''',
    )

How to run tests

Unit tests

From the root project directory: python -m unittest discover -s tests/unit

Integration tests

Integration tests require an access to ClickHouse server. Tests use connection URI defined via environment variable AIRFLOW_CONN_CLICKHOUSE_DEFAULT with clickhouse://localhost as default.

Run from the project root: python -m unittest discover -s tests/integration

All tests

From the root project directory: python -m unittest discover -s tests

Contributors

Anton Bryzgalov, @bryzgaloff
Viktor Taranenko, @viktortnk
Danila Ganchar, @d-ganchar
Mikhail, @glader
Alexander Chashnikov, @ne1r0n

arturkuchynski/airflow-clickhouse-plugin

Airflow ClickHouse Plugin

Features

Installation and dependencies

Usage

ClickHouseOperator Reference

ClickHouseHook Reference

ClickHouseSqlSensor Reference

ClickHouse Connection schema

Extra arguments

Default values

Default connection

Examples

ClickHouseOperator Example

ClickHouseHook Example

ClickHouseSqlSensor Example

How to run tests

Unit tests

Integration tests

All tests

Contributors