Skip to main content

Treasure Data API library for Python

Project description

Build Status on GitHub ACtions Build status Coverage Status PyPI version

Treasure Data API library for Python


td-client supports the following versions of Python.

  • Python 3.5+

  • PyPy


You can install the releases from PyPI.

$ pip install td-client

It’d be better to install certifi to enable SSL certificate verification.

$ pip install certifi


Please see also the examples at Treasure Data Documentation.

The td-client documentation is hosted at, or you can go directly to the API documentation.

For information on the parameters that may be used when reading particular types of data, see File import parameters.

Listing jobs

Treasure Data API key will be read from environment variable TD_API_KEY, if none is given via apikey= argument passed to tdclient.Client.

Treasure Data API endpoint is used by default. You can override this with environment variable TD_API_SERVER, which in turn can be overridden via endpoint= argument passed to tdclient.Client. List of available Treasure Data sites and corresponding API endpoints can be found here.

import tdclient

with tdclient.Client() as td:
    for job in

Running jobs

Running jobs on Treasure Data.

import tdclient

with tdclient.Client() as td:
    job = td.query("sample_datasets", "SELECT COUNT(1) FROM www_access", type="hive")
    for row in job.result():

Running jobs via DBAPI2

td-client-python implements PEP 0249 Python Database API v2.0. You can use td-client-python with external libraries which supports Database API such like pandas.

import pandas
import tdclient

def on_waiting(cursor):

with tdclient.connect(db="sample_datasets", type="presto", wait_callback=on_waiting) as td:
    data = pandas.read_sql("SELECT symbol, COUNT(1) AS c FROM nasdaq GROUP BY symbol", td)

We offer another package for pandas named pytd with some advanced features. You may prefer it if you need to do complicated things, such like exporting result data to Treasure Data, printing job’s progress during long execution, etc.

Importing data

Importing data into Treasure Data in streaming manner, as similar as fluentd is doing.

import sys
import tdclient

with tdclient.Client() as td:
    for file_name in sys.argv[:1]:
        td.import_file("mydb", "mytbl", "csv", file_name)

Bulk import

Importing data into Treasure Data in batch manner.

import sys
import tdclient
import uuid
import warnings

if len(sys.argv) <= 1:

with tdclient.Client() as td:
    session_name = "session-{}".format(uuid.uuid1())
    bulk_import = td.create_bulk_import(session_name, "mydb", "mytbl")
        for file_name in sys.argv[1:]:
            part_name = "part-{}".format(file_name)
            bulk_import.upload_file(part_name, "json", file_name)
    if 0 < bulk_import.error_records:
        warnings.warn("detected {} error records.".format(bulk_import.error_records))
    if 0 < bulk_import.valid_records:
        print("imported {} records.".format(bulk_import.valid_records))
        raise(RuntimeError("no records have been imported: {}".format(

If you want to import data as msgpack format, you can write as follows:

import io
import time
import uuid
import warnings

import tdclient

t1 = int(time.time())
l1 = [{"a": 1, "b": 2, "time": t1}, {"a": 3, "b": 9, "time": t1}]

with tdclient.Client() as td:
    session_name = "session-{}".format(uuid.uuid1())
    bulk_import = td.create_bulk_import(session_name, "mydb", "mytbl")
        _bytes = tdclient.util.create_msgpack(l1)
        bulk_import.upload_file("part", "msgpack", io.BytesIO(_bytes))
    # same as the above example

Changing how CSV and TSV columns are read

The td-client package will generally make sensible choices on how to read the columns in CSV and TSV data, but sometimes the user needs to override the default mechanism. This can be done using the optional file import parameters dtypes and converters.

For instance, consider CSV data that starts with the following records:


If that data is read using the defaults, it will produce values that look like:

1575454204, "a", 1, "a;b;c"
1575454204, "b", 2, "d;e;f"

that is, an integer, a string, an integer and another string.

If the user wants to keep the leading zeroes in col2, then they can specify the column datatype as string. For instance, using bulk_import.upload_file to read data from input_data:

    "part", "msgpack", input_data,
    dtypes={"col2": "str"},

which would produce:

1575454204, "a", "0001", "a;b;c"
1575454204, "b", "0002", "d;e;f"

If they also wanted to treat col3 as a sequence of strings, separated by semicolons, then they could specify a function to process col3:

    "part", "msgpack", input_data,
    dtypes={"col2": "str"},
    converters={"col3", lambda x: x.split(";")},

which would produce:

1575454204, "a", "0001", ["a", "b", "c"]
1575454204, "b", "0002", ["d", "e", "f"]


Running tests

Run tests.

$ python test

Running tests (tox)

You can run tests against all supported Python versions. I’d recommend you to install pyenv to manage Pythons.

$ pyenv shell system
$ for version in $(cat .python-version); do [ -d "$(pyenv root)/versions/${version}" ] || pyenv install "${version}"; done
$ pyenv shell --unset

Install tox.

$ pip install tox

Then, run tox.

$ tox


Release to PyPI. Ensure you installed twine.

$ python bdist_wheel sdist
$ twine upload dist/*


Apache Software License, Version 2.0

Project details

Release history Release notifications | RSS feed

This version


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

td-client-1.2.1.tar.gz (60.6 kB view hashes)

Uploaded Source

Built Distribution

td_client-1.2.1-py3-none-any.whl (86.1 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page