Data Loading and Input

Before data can be queried with dask-sql, it needs to be loaded into the Dask cluster (or local instance) and registered with the Context. dask-sql supports all dask-compatible input formats, plus some additional formats only suitable for dask-sql.

1. Load it via Python

You can either use already created Dask DataFrames or create one by using the create_table() function. Chances are high, there exists already a function to load your favorite format or location (e.g. S3 or hdfs). See below for all formats understood by dask-sql. Make sure to install required libraries both on the driver and worker machines:

import dask.dataframe as dd
from dask_sql import Context

c = Context()
df = dd.read_csv("s3://nyc-tlc/trip data/yellow_tripdata_2019-01.csv")

c.create_table("my_data", df)

or in short (equivalent):

from dask_sql import Context

c = Context()

c.create_table("my_data", "s3://nyc-tlc/trip data/yellow_tripdata_2019-01.csv")

2. Load it via SQL

If you are connected to the SQL server implementation or you do not want to issue Python command calls, you can also achieve the data loading via SQL only.

    format = 'csv',
    location = 's3://nyc-tlc/trip data/yellow_tripdata_2019-01.csv'

The parameters are the same as in the Python function described above. You can find more information in Table Creation.

3. Persist and share data on the cluster

In dask, you can publish datasets with names into the cluster memory. This allows to reuse the same data from multiple clients/users in multiple sessions.

For example, you can publish your data using the client.publish_dataset function of the distributed.Client, and then later register it in the Context via SQL:

# a dask.distributed Client
client = Client(...)

Later in SQL:

    format = 'memory',
    location = 'my_df'

Note, that the format is set to memory and the location is the name, which was chosen when publishing the dataset.

To achieve the same thing from Python, you can just use Dask’s methods to get the dataset

df = client.get_dataset("my_df")
c.create_table("my_data", df)

Input Formats

dask-sql understands (thanks to the large Dask ecosystem) a wide verity of input formats and input locations.

  • All formats and locations mentioned in the Dask documentation, including CSV, Parquet, and JSON. Just pass in the location as string (and possibly the format, e.g. “csv” if it is not clear from the file extension). The data can be from local disc or many remote locations (S3, hdfs, Azure Filesystem, http, Google Filesystem, …) - just prefix the path with the matching protocol. Additional arguments passed to create_table() or CREATE TABLE are given to the read_<format> calls.


    storage_options={'anon': True}
    format = 'csv', -- can also be omitted, as clear from the extension
    location = 's3://bucket-name/my-data-*.csv',
    storage_options = (
        anon = True
  • If your data is already in Pandas (or Dask) DataFrames format, you can just use it as it is via the Python API by giving it to create_table() directly.

  • You can connect dask-sql to an intake catalog and use the data registered there. Assuming you have an intake catalog stored in “catalog.yaml” (can also be the URL of an intake server), you can read in a stored table “data_table” either via Python

    catalog = intake.open_catalog("catalog.yaml")
    c.create_table("my_data", catalog, intake_table_name="intake_table")
    # or
    c.create_table("my_data", "catalog.yaml", format="intake", intake_table_name="intake_table")

    or via SQL:

    CREATE TABLE my_data WITH (
        format = 'intake',
        location = 'catalog.yaml'

    The argument intake_table_name is optional and defaults to the table name in dask_sql. With the argument catalog_kwargs you can control how the intake catalog object is created. Additional arguments are forwarded to the to_dask() call of intake.

  • As an experimental feature, it is also possible to use data stored in the Apache Hive metastore. For this, dask-sql will retrieve the information on the storage location and format from the metastore and will then register the raw data directly in the context. This means, no Hive data query will be issued and you might be able to see a speed improvement.

    It is both possible to use a pyhive.hive.Cursor or an sqlalchemy connection.

    from dask_sql import Context
    from pyhive.hive import connect
    import sqlalchemy
    c = Context()
    cursor = connect("hive-server", 10000).cursor()
    # or
    cursor = sqlalchemy.create_engine("hive://hive-server:10000").connect()
    c.create_table("my_data", cursor, hive_table_name="the_name_in_hive")

    or in SQL:

    CREATE TABLE my_data WITH (
        location = 'hive://hive-server:10000',
        hive_table_name = 'the_name_in_hive'

    Again, hive_table_name is optional and defaults to the table name in dask-sql. You can also control the database used in Hive via the hive_schema_name parameter. Additional arguments are pushed to the internally called read_<format> functions.

  • Similarly, it is possible to load data from a Databricks Cluster (which is similar to a Hive metastore).

    You need to have the databricks-dbapi package installed and fsspec >= 0.8.7. A token needs to be generated for the accessing user. The host, port and http_path information can be found in the JDBC tab of the cluster.

    from dask_sql import Context
    from sqlalchemy import create_engine
    c = Context()
    cursor = create_engine(f"databricks+pyhive://token:{token}@{host}:{port}/",
                           connect_args={"http_path": http_path}).connect()
    c.create_table("my_data", cursor, hive_table_name="schema.table",
                   storage_options={"instance": host, "token": token})

    or in SQL

    CREATE TABLE my_data WITH (
        location = 'databricks+pyhive://token:{token}@{host}:{port}/',
        connect_args = (
            http_path = '{http_path}'
        hive_table_name = 'schema.table',
        storage_options = (
            instance = '{host}',
            token = '{token}'


For dask-sql it does not matter how you load your data. In all shown cases you can then use the specified table name to query your data in a SELECT call.

Please note however that un-persisted data will be reread from its source (e.g. on S3 or disk) on every query whereas persisted data is only read once. This will increase the query speed, but will also prevent you from seeing external updates to your data (until you reload it explicitly).