Skip to main content

...About 7 min

Python Native API


You have to install thrift (>=0.13) before using the package.

How to use (Example)

First, download the package: pip3 install apache-iotdb

You can get an example of using the package to read and write data at here: Exampleopen in new window

An example of aligned timeseries: Aligned Timeseries Session Exampleopen in new window

(you need to add import iotdb in the head of the file)


from iotdb.Session import Session

ip = ""
port_ = "6667"
username_ = "root"
password_ = "root"
session = Session(ip, port_, username_, password_)
zone = session.get_time_zone()


  • Initialize a Session
session = Session(ip, port_, username_, password_, fetch_size=1024, zone_id="UTC+8")
  • Open a session, with a parameter to specify whether to enable RPC compression

Notice: this RPC compression status of client must comply with that of IoTDB server

  • Close a Session

Data Definition Interface (DDL Interface)

Database Management

  • Delete one or several databases

Timeseries Management

  • Create one or multiple timeseries
session.create_time_series(ts_path, data_type, encoding, compressor,
    props=None, tags=None, attributes=None, alias=None)
    ts_path_lst, data_type_lst, encoding_lst, compressor_lst,
    props_lst=None, tags_lst=None, attributes_lst=None, alias_lst=None
  • Create aligned timeseries
    device_id, measurements_lst, data_type_lst, encoding_lst, compressor_lst

Attention: Alias of measurements are not supported currently.

  • Delete one or several timeseries
  • Check whether the specific timeseries exists

Data Manipulation Interface (DML Interface)


It is recommended to use insertTablet to help improve write efficiency.

  • Insert a Tablet,which is multiple rows of a device, each row has the same measurements
    • Better Write Performance
    • Support null values: fill the null value with any value, and then mark the null value via BitMap (from v0.13)

We have two implementations of Tablet in Python API.

  • Normal Tablet
values_ = [
    [False, 10, 11, 1.1, 10011.1, "test01"],
    [True, 100, 11111, 1.25, 101.0, "test02"],
    [False, 100, 1, 188.1, 688.25, "test03"],
    [True, 0, 0, 0, 6.25, "test04"],
timestamps_ = [1, 2, 3, 4]
tablet_ = Tablet(
    device_id, measurements_, data_types_, values_, timestamps_
  • Numpy Tablet

Comparing with Tablet, Numpy Tablet is using numpy.ndarrayopen in new window to record data.
With less memory footprint and time cost of serialization, the insert performance will be better.


  1. time and numerical value columns in Tablet is ndarray
  2. recommended to use the specific dtypes to each ndarray, see the example below
    (if not, the default dtypes are also ok).
import numpy as np
data_types_ = [
np_values_ = [
    np.array([False, True, False, True], TSDataType.BOOLEAN.np_dtype()),
    np.array([10, 100, 100, 0], TSDataType.INT32.np_dtype()),
    np.array([11, 11111, 1, 0], TSDataType.INT64.np_dtype()),
    np.array([1.1, 1.25, 188.1, 0], TSDataType.FLOAT.np_dtype()),
    np.array([10011.1, 101.0, 688.25, 6.25], TSDataType.DOUBLE.np_dtype()),
    np.array(["test01", "test02", "test03", "test04"]),
np_timestamps_ = np.array([1, 2, 3, 4], TSDataType.INT64.np_dtype())
np_tablet_ = NumpyTablet(
  "root.sg_test_01.d_02", measurements_, data_types_, np_values_, np_timestamps_
  • Insert multiple Tablets
  • Insert a Record
session.insert_record(device_id, timestamp, measurements_, data_types_, values_)
  • Insert multiple Records
    device_ids_, time_list_, measurements_list_, data_type_list_, values_list_
  • Insert multiple Records that belong to the same device.
    With type info the server has no need to do type inference, which leads a better performance
session.insert_records_of_one_device(device_id, time_list, measurements_list, data_types_list, values_list)

Insert with type inference

When the data is of String type, we can use the following interface to perform type inference based on the value of the value itself. For example, if value is "true" , it can be automatically inferred to be a boolean type. If value is "3.2" , it can be automatically inferred as a flout type. Without type information, server has to do type inference, which may cost some time.

  • Insert a Record, which contains multiple measurement value of a device at a timestamp
session.insert_str_record(device_id, timestamp, measurements, string_values)

Insert of Aligned Timeseries

The Insert of aligned timeseries uses interfaces like insert_aligned_XXX, and others are similar to the above interfaces:

  • insert_aligned_record
  • insert_aligned_records
  • insert_aligned_records_of_one_device
  • insert_aligned_tablet
  • insert_aligned_tablets

IoTDB-SQL Interface

  • Execute query statement
  • Execute non query statement
  • Execute statement

Schema Template

Create Schema Template

The step for creating a schema template is as follows

  1. Create the template class
  2. Adding child Node,InternalNode and MeasurementNode can be chose
  3. Execute create schema template function
template = Template(name=template_name, share_time=True)

i_node_gps = InternalNode(name="GPS", share_time=False)
i_node_v = InternalNode(name="vehicle", share_time=True)
m_node_x = MeasurementNode("x", TSDataType.FLOAT, TSEncoding.RLE, Compressor.SNAPPY)




Modify Schema Template measurements

Modify measurements in a template, the template must be already created. These are functions that add or delete some measurement nodes.

  • add node in template
session.add_measurements_in_template(template_name, measurements_path, data_types, encodings, compressors, is_aligned)
  • delete node in template
session.delete_node_in_template(template_name, path)

Set Schema Template

session.set_schema_template(template_name, prefix_path)

Uset Schema Template

session.unset_schema_template(template_name, prefix_path)

Show Schema Template

  • Show all schema templates
  • Count all measurements in templates
  • Judge whether the path is measurement or not in templates, This measurement must be in the template
session.count_measurements_in_template(template_name, path)
  • Judge whether the path is exist or not in templates, This path may not belong to the template
session.is_path_exist_in_template(template_name, path)
  • Show nodes under in schema template
  • Show the path prefix where a schema template is set
  • Show the path prefix where a schema template is used (i.e. the time series has been created)

Drop Schema Template

Delete an existing schema template,dropping an already set template is not supported


Pandas Support

To easily transform a query result to a Pandas Dataframeopen in new window
the SessionDataSet has a method .todf() which consumes the dataset and transforms it to a pandas dataframe.


from iotdb.Session import Session

ip = ""
port_ = "6667"
username_ = "root"
password_ = "root"
session = Session(ip, port_, username_, password_)
result = session.execute_query_statement("SELECT * FROM root.*")

# Transform to Pandas Dataset
df = result.todf()


# Now you can work with the dataframe
df = ...

IoTDB Testcontainer

The Test Support is based on the lib testcontainers ( in new window) which you need to install in your project if you want to use the feature.

To start (and stop) an IoTDB Database in a Docker container simply do:

class MyTestCase(unittest.TestCase):

    def test_something(self):
        with IoTDBContainer() as c:
            session = Session("localhost", c.get_exposed_port(6667), "root", "root")
            result = session.execute_query_statement("SHOW TIMESERIES")

by default it will load the image apache/iotdb:latest, if you want a specific version just pass it like e.g. IoTDBContainer("apache/iotdb:0.12.0") to get version 0.12.0 running.


IoTDB DBAPI implements the Python DB API 2.0 specification ( in new window), which defines a common
interface for accessing databases in Python.


  • Initialization

The initialized parameters are consistent with the session part (except for the sqlalchemy_mode).

from iotdb.dbapi import connect

ip = ""
port_ = "6667"
username_ = "root"
password_ = "root"
conn = connect(ip, port_, username_, password_,fetch_size=1024,zone_id="UTC+8",sqlalchemy_mode=False)
cursor = conn.cursor()
  • simple SQL statement execution
cursor.execute("SELECT ** FROM root")
for row in cursor.fetchall():
  • execute SQL with parameter

IoTDB DBAPI supports pyformat style parameters

cursor.execute("SELECT ** FROM root WHERE time < %(time)s",{"time":"2017-11-01T00:08:00.000"})
for row in cursor.fetchall():
  • execute SQL with parameter sequences
seq_of_parameters = [
    {"timestamp": 1, "temperature": 1},
    {"timestamp": 2, "temperature": 2},
    {"timestamp": 3, "temperature": 3},
    {"timestamp": 4, "temperature": 4},
    {"timestamp": 5, "temperature": 5},
sql = "insert into root.cursor(timestamp,temperature) values(%(timestamp)s,%(temperature)s)"
  • close the connection and cursor

IoTDB SQLAlchemy Dialect (Experimental)

The SQLAlchemy dialect of IoTDB is written to adapt to Apache Superset.
This part is still being improved.
Please do not use it in the production environment!

Mapping of the metadata

The data model used by SQLAlchemy is a relational data model, which describes the relationships between different entities through tables.
While the data model of IoTDB is a hierarchical data model, which organizes the data through a tree structure.
In order to adapt IoTDB to the dialect of SQLAlchemy, the original data model in IoTDB needs to be reorganized.
Converting the data model of IoTDB into the data model of SQLAlchemy.

The metadata in the IoTDB are:

  1. Database
  2. Path
  3. Entity
  4. Measurement

The metadata in the SQLAlchemy are:

  1. Schema
  2. Table
  3. Column

The mapping relationship between them is:

The metadata in the SQLAlchemyThe metadata in the IoTDB
TablePath ( from database to entity ) + Entity

The following figure shows the relationship between the two more intuitively:


Data type mapping

data type in IoTDBdata type in SQLAlchemy


  • execute statement
from sqlalchemy import create_engine

engine = create_engine("iotdb://root:root@")
connect = engine.connect()
result = connect.execute("SELECT ** FROM root")
for row in result.fetchall():
  • ORM (now only simple queries are supported)
from sqlalchemy import create_engine, Column, Float, BigInteger, MetaData
from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy.orm import sessionmaker

metadata = MetaData(
Base = declarative_base(metadata=metadata)

class Device(Base):
    __tablename__ = "room2.device1"
    Time = Column(BigInteger, primary_key=True)
    temperature = Column(Float)
    status = Column(Float)

engine = create_engine("iotdb://root:root@")

DbSession = sessionmaker(bind=engine)
session = DbSession()

res = session.query(Device.status).filter(Device.temperature > 1)

for row in res:



This is an example of how to connect to IoTDB with python, using the thrift rpc interfaces. Things are almost the same on Windows or Linux, but pay attention to the difference like path separator.


Python3.7 or later is preferred.

You have to install Thrift (0.11.0 or later) to compile our thrift file into python code. Below is the official tutorial of installation, eventually, you should have a thrift executable.

Before starting you need to install requirements_dev.txt in your python environment, e.g. by calling

pip install -r requirements_dev.txt

Compile the thrift library and Debug

In the root of IoTDB's source code folder, run mvn clean generate-sources -pl client-py -am.

This will automatically delete and repopulate the folder iotdb/thrift with the generated thrift files.
This folder is ignored from git and should never be pushed to git!

Notice Do not upload iotdb/thrift to the git repo.

Session Client & Example

We packed up the Thrift interface in client-py/src/iotdb/ (similar with its Java counterpart), also provided an example file client-py/src/ of how to use the session module. please read it carefully.

Or, another simple example:

from iotdb.Session import Session

ip = ""
port_ = "6667"
username_ = "root"
password_ = "root"
session = Session(ip, port_, username_, password_)
zone = session.get_time_zone()


Please add your custom tests in tests folder.

To run all defined tests just type pytest . in the root folder.

Notice Some tests need docker to be started on your system as a test instance is started in a docker container using testcontainersopen in new window.

Futher Tools

blackopen in new window and flake8open in new window are installed for autoformatting and linting.
Both can be run by black . or flake8 . respectively.


To do a release just ensure that you have the right set of generated thrift files.
Then run linting and auto-formatting.
Then, ensure that all tests work (via pytest .).
Then you are good to go to do a release!

Preparing your environment

First, install all necessary dev dependencies via pip install -r requirements_dev.txt.

Doing the Release

There is a convenient script to do all steps for a release.
Namely, these are

  • Remove all transient directories from last release (if exists)
  • (Re-)generate all generated sources via mvn
  • Run Linting (flake8)
  • Run Tests via pytest
  • Build
  • Release to pypi

Copyright © 2024 The Apache Software Foundation.
Apache and the Apache feather logo are trademarks of The Apache Software Foundation

Have a question? Connect with us on QQ, WeChat, or Slack. Join the community now.