Skip to main content

Dynatrace OneAgent SDK for Python

Project description

Read the latest version of this README, with working internal links, at GitHub.

Dynatrace OneAgent SDK for Python

This SDK enables Dynatrace customers to extend request level visibility into Python applications. It provides the Python implementation of the Dynatrace OneAgent SDK.


The SDK supports Python 2 ≥ 2.7 and Python 3 ≥ 3.4. Only the official CPython (that is, the "normal" Python, i.e. the Python implementation from is supported and only on Linux (musl libc which is used, e.g., on Alpine Linux, is currently not supported) and Windows with the x86 (including x86-64) architecture. Additionally, pip ≥ 8.1.0 (2016-03-05) is required for installation, and on Linux, the system should be manylinux1-compatible to ensure a smooth installation via pip.

The Dynatrace OneAgent SDK for Python is a wrapper of the Dynatrace OneAgent SDK for C/C++ and therefore the SDK for C/C++ is required and delivered with the Python SDK. See here for its requirements, which also apply to the SDK for Python.

The version of the SDK for C/C++ that is included in each version of the SDK for Python is shown in the following table along with the required Dynatrace OneAgent version (it is the same as listed in the OneAgent SDK for C/C++'s documentation).

OneAgent SDK for Python OneAgent SDK for C/C++ Dynatrace OneAgent Support status
1.4 1.6.1 ≥1.179 Supported
1.3 1.5.1 ≥1.179 Supported
1.2 1.4.1 ≥1.161 Deprecated¹
1.1 1.3.1 ≥1.151 Deprecated¹
1.0 1.1.0 ≥1.141 EAP (not supported)
  1. Deprecated releases of the OneAgent SDK for Python are still supported but this might change with a future release. Applications using those deprecated versions should be upgraded to the latest release.

Using the OneAgent SDK for Python in your application

To install the latest version of the OneAgent SDK for Python, use the PyPI package oneagent-sdk:

python -m pip install --upgrade oneagent-sdk

To verify your installation, execute

python -c "import oneagent; print(oneagent.initialize())"

If the installation was successful, you should get an output ending with InitResult(status=0, error=None). Otherwise, see the Troubleshooting section.

To load the OneAgent SDK into your application, just add the following line at the top of your script:

import oneagent

Here is a quick "Hello World" that will produce a service call in Dynatrace:

import oneagent

if not oneagent.initialize():
    print('Error initializing OneAgent SDK.')

with oneagent.get_sdk().trace_incoming_remote_call('method', 'service', 'endpoint'):

print('It may take a few moments before the path appears in the UI.')
input('Please wait...')

A more detailed sample application is available here. See also the Quickstart section in the API documentation.

API Concepts

Common concepts of the Dynatrace OneAgent SDK are explained in the Dynatrace OneAgent SDK repository.

Initialization and SDK objects

Before first using any other SDK functions, you need to initialize the SDK.

init_result = oneagent.initialize()
print('OneAgent SDK initialization result' + repr(init_result))
if init_result:
    print('SDK should work (but agent might be inactive).')
    print('SDK will definitely not work (i.e. functions will be no-ops):', init_result)

See the API documentation for the initialize function and the InitResult class for more information.

To use the SDK, get a reference to the SDK singleton by calling the oneagent static get_sdk method. The first thing you may want to do with this object, is checking if the agent is active by comparing the value of the agent_state property to the oneagent.common.AgentState constants. You can also have a look at the extended SDK state information.

import oneagent
from oneagent.common import AgentState
# Initialize oneagent, as above

sdk = oneagent.get_sdk()
if sdk.agent_state not in (AgentState.ACTIVE, AgentState.TEMPORARILY_INACTIVE):
    print('Too bad, you will not see data from this process.')

As a development and debugging aid it is recommended to set a diagnostic callback. The callback will be used by the SDK to inform about unusual events.

Unusual events that prevent an operation from completing successfully include:

  • API usage errors
  • other unexpected events (like out of memory situations)

NOTE: Use this as a development and debugging aid only. Your application should not rely on a calling sequence or any message content being set or passed to the callback.

During development, it is additionally recommended to use the "verbose callback" which also informs about other events that may be benign but can be very helpful in debugging, e.g. a PurePath that was not created because a Tracer is disabled by configuration, etc.

def _diag_callback(unicode_message):

sdk.set_verbose_callback(_diag_callback) # Do not use this callback in production


To trace any kind of call you first need to create a Tracer, using one of the various trace_* methods of the SDK object. The Tracer object controls the “life cycle” of a trace: Entering a with-block with a tracer starts the trace, exiting it ends it. Exiting the with block with an exception causes the trace to be marked as failed with the exception message (if you do not want or need this behavior, tracers have explicit methods for starting, ending and attaching error information too; see the documentation).

There are different tracer types requiring different information for creation. As an example, to trace an incoming remote call, this would be the most simple way to trace it:

import oneagent

with oneagent.get_sdk().trace_incoming_remote_call('method', 'service', 'endpoint'):
    pass # Here you would do the actual work that is timed

See the section on remote calls for more information.

Some tracers also support attaching additional information before ending it.

Important: In Python 2, tracers accept both byte (“normal”) and unicode strings. Byte strings must always use the UTF-8 encoding!

Features and how to use them

The feature sets differ slightly with each language implementation. More functionality will be added over time, see Planned features for OneAgent SDK for details on upcoming features.

A more detailed specification of the features can be found in Dynatrace OneAgent SDK.

Feature Required OneAgent SDK for Python version
Custom services ≥1.2.0
Messaging ≥1.2.0
In-process linking ≥1.1.0
Custom request attributes ≥1.1.0
Outgoing web requests ≥1.1.0
Incoming web requests ≥1.0.0
SQL database requests ≥1.0.0
Trace incoming and outgoing remote calls ≥1.0.0

Remote calls

You can use the SDK to trace communication from one process to another. This will enable you to see full Service Flow, PurePath and Smartscape topology for remoting technologies that Dynatrace is not aware of.

To trace any kind of remote call you first need to create a Tracer. The Tracer object represents the endpoint that you want to call, thus you need to supply the name of the remote service and method. In addition, you need to transport a tag in your remote call from the client side to the server side if you want to trace it end to end.

On the client side, you would trace the outgoing remote call like this:

outcall = sdk.trace_outgoing_remote_call(
    'remoteMethodToCall', 'RemoteServiceName', 'rmi://Endpoint/service',
    oneagent.sdk.Channel(oneagent.sdk.ChannelType.TCP_IP, 'remoteHost:1234'),
with outcall:
    # Note: You can access outgoing_dynatrace_*_tag only after the trace
    # has started!
    strtag = outcall.outgoing_dynatrace_string_tag
    do_actual_remote_call(extra_headers={'X-dynaTrace': strtag})

On the server side, you would trace it like this:

incall = sdk.trace_incoming_remote_call(
    'remoteMethodToCall', 'RemoteServiceName', 'rmi://Endpoint/service',
with incall:
    pass # Here you would do the actual work that is timed

See the documentation for more information:

SQL database requests

To trace database requests you need a database info object which stores the information about your database which does not change between individual requests. This will typically be created somewhere in your initialization code (after initializing the SDK):

dbinfo = sdk.create_database_info(
    'Northwind', oneagent.sdk.DatabaseVendor.SQLSERVER,
    oneagent.sdk.Channel(oneagent.sdk.ChannelType.TCP_IP, ''))

Then you can trace the SQL database requests:

with sdk.trace_sql_database_request(dbinfo, 'SELECT foo FROM bar;') as tracer:
    # Do actual DB request
    tracer.set_rows_returned(42) # Optional
    tracer.set_round_trip_count(3) # Optional

Note that you need to release the database info object. You can do this by calling close() on it or using it in a with block.

See the documentation for more information:

Please note that SQL database traces are only created if they occur within some other SDK trace (e.g. incoming remote call).

Incoming web requests

Same as with database infos, to trace incoming web requests you need a web application info object which stores the information about your web application which does not change:

wappinfo = sdk.create_web_application_info(

Then you can trace incoming web requests:

wreq = sdk.trace_incoming_web_request(
    headers={'Host': '', 'X-foo': 'bar'},

with wreq:
    wreq.add_parameter('my_form_field', '1234')
    # Process web request
    wreq.add_response_headers({'Content-Length': '1234'})
    wreq.set_status_code(200) # OK

Note that you need to release the web application info object. You can do this by calling close() on it or using it in a with block.

Incoming web request tracers support some more features not shown here. Be sure to check out the documentation:

Outgoing web requests

To trace an outgoing web request you need to create an 'Outgoing Web Request Tracer' object. You pass the destination URL, the HTTP method and request headers as parameters.

Let's have a look at a web request example:

from urllib.request import Request

# Create your web request.
url = ''

req = Request(url)
req.add_header('header1', '1234')
req.add_header('header2', '5678')

After creating/setting up the request you have to create the tracer object and pass the parameters.

# Create the tracer.
tracer = sdk.trace_outgoing_web_request(url, req.get_method(), req.headers)

The next step is to start the tracer and then to retrieve the outgoing Dynatrace tag. The tag is being used to trace a transaction from end-to-end. You have to send the tag to the destination via an additional request header which is called DYNATRACE_HTTP_HEADER_NAME. Here you can find more information on tagging.

with tracer:
	# Get and set the Dynatrace tag.
	tag = tracer.outgoing_dynatrace_string_tag
 	req.add_header(DYNATRACE_HTTP_HEADER_NAME, tag)

	# Here you process and send the web request.
	response = _process_your_outgoing_request(req)

Finally, get the response headers you want to trace and the status code of the response and add them to the tracer.

        tracer.add_response_headers({'Content-Length': response.get_content_length()})

Be sure to check out the documentation:

Trace in-process asynchronous execution

You can use the SDK to trace asynchronous in-process code execution. This might be useful if the OneAgent does not support the threading framework or specific asynchronous libraries. In-process linking should be used to link other services (Database, Webrequests, ...) between thread or queueing boundaries currently not supported out-of-the-box by the OneAgent.

To link asynchronous execution, you need to create an in-process link, where the execution forks:

in_process_link = sdk.create_in_process_link()

The provided in-process link must not be serialized and can only be used inside the process in which it was created. It must be used to start tracing where the asynchronous execution takes place:

with sdk.trace_in_process_link(in_process_link):
 	# Do the asynchronous job

Custom Request Attributes

You can use the SDK to add custom request attributes to the currently traced service. Custom request attributes allow you to do easier/better filtering of your requests in Dynatrace.

Adding custom request attributes to the currently traced service call is pretty simple. Just call the add_custom_request_attribute method with your key and value (only int, float and string values are currently supported):

sdk.add_custom_request_attribute('errorCount', 42)
sdk.add_custom_request_attribute('gross weight', 2.39)
sdk.add_custom_request_attribute('famous actor', 'Benedict Cumberbatch')

Check out the documentation at:

Custom services

You can use the SDK to trace custom service methods. A custom service method is a meaningful part of your code that you want to trace but that does not fit any other tracer. An example could be the callback of a periodic timer.

with sdk.trace_custom_service('onTimer', 'CleanupTask'):
	# Do the cleanup task

Check out the documentation at:


You can use the SDK to trace messages sent or received via a messaging system. When tracing messages, we distinguish between:

  • sending a message
  • waiting for and receiving a message
  • processing a received message

Outgoing Messages

All messaging related tracers need a messaging system info object which you have to create prior to the respective messaging tracer, which is an outgoing message tracer in the example below.

msi_handle = sdk.create_messaging_system_info(
	'myMessagingSystem', 'requestQueue', MessagingDestinationType.QUEUE,
	oneagent.sdk.Channel(oneagent.sdk.ChannelType.TCP_IP, ''))

with msi_handle:
	with sdk.trace_outgoing_message(msi_handle) as tracer:
		# Get and set the Dynatrace tag.
		tag = tracer.outgoing_dynatrace_string_tag
		message_to_send.add_header_field(oneagent.sdk.DYNATRACE_MESSAGE_PROPERTY_NAME, tag)

		# Send the message.

		# Optionally set message and/or correlation IDs

Incoming Messages

On the incoming side, we need to differentiate between the blocking receiving part and processing the received message. Therefore two different tracers are being used:

  • IncomingMessageReceiveTracer
  • IncomingMessageProcessTracer
msi_handle = sdk.create_messaging_system_info(
	'myMessagingSystem', 'requestQueue', MessagingDestinationType.QUEUE,
	oneagent.sdk.Channel(oneagent.sdk.ChannelType.TCP_IP, ''))

with msi_handle:
	# Create the receive tracer for incoming messages.
	with sdk.trace_incoming_message_receive(msi_handle):
		# This is a blocking call, which will return as soon as a message is available.
		Message query_message = the_queue.receive()

		# Get the Dynatrace tag from the message.
		tag = query_message.get_header_field(oneagent.sdk.DYNATRACE_MESSAGE_PROPERTY_NAME)

		# Create the tracer for processing incoming messages.
		tracer = sdk.trace_incoming_message_process(msi_handle, str_tag=tag)

		with tracer:
			# Now let's handle the message ...
			print('handle incoming message')

In case of non-blocking receive (e. g. using an event handler), there is no need to use an IncomingMessageReceiveTracer - just trace processing of the message by using the IncomingMessageProcessTracer:

msi_handle = sdk.create_messaging_system_info(
	'myMessagingSystem', 'requestQueue', MessagingDestinationType.QUEUE,
	oneagent.sdk.Channel(oneagent.sdk.ChannelType.TCP_IP, ''))

def on_message_received(message):
	# Get the Dynatrace tag from the message.
	tag = message.get_header_field(oneagent.sdk.DYNATRACE_MESSAGE_PROPERTY_NAME)

	# Create the tracer for processing incoming messages.
	tracer = sdk.trace_incoming_message_process(msi_handle, str_tag=tag)

	with tracer:
		# Now let's handle the message ...
		print('handle incoming message')

See the documentation for more information:

Using the OneAgent SDK for Python with forked child processes (only available on Linux)

Some applications, especially web servers, use a concurrency model that is based on forked child processes. Typically a master process is started which is responsible only for creating and managing child processes by means of forking. The child processes do the real work, for example handling web requests.

The recommended way to use the Python SDK in such a scenario is as follows: You initialize the SDK in the master process setting the forkable argument to True.

oneagent.initialize(sdk_options, forkable=True)

This way you will not be able to use the SDK in the master process (attempts to do so will be ignored, if applicable with an error code), but all forked child processes will share the same agent. This has a lower overhead, for example the startup of worker processes is not slowed down, and the per-worker memory overhead is reduced.

For more information on forked child processes, take a look at those resources:


To debug your OneAgent SDK for Python installation, execute the following Python code:

import logging
import time
import oneagent

log_handler = logging.StreamHandler()
log_formatter = logging.Formatter(
    '%(asctime)s.%(msecs)03d UTC [%(thread)08x]'
    ' %(levelname)-7s [%(name)-6s] %(message)s',
    '%Y-%m-%d %H:%M:%S')
log_formatter.converter = time.gmtime
init_result = oneagent.initialize(['loglevelsdk=finest', 'loglevel=finest'])
print('InitResult=' + repr(init_result))

If you get output containing InitResult=InitResult(status=0, error=None), your installation should be fine. Otherwise, the output is helpful in determining the issue. The extended SKD state might also help to diagnose your problem.

Known gotchas:

  • ImportError or ModuleNotFoundError in line 1 that says that there is no module named oneagent.

    Make sure that the pip install or equivalent succeeded (see here). Also make sure you use the pip corresponding to your python (if in doubt, use python -m pip instead of pip for installing).

  • Output ending in a message like InitResult=InitResult(status=-2, error=SDKError(-1342308345, 'Failed loading SDK stub from .../site-packages/oneagent/_impl/native/ "/.../ cannot open shared object file: No such file or directory". Check your installation of the oneagent-sdk Python package, e.g., try running pip install --verbose --force-reinstall oneagent-sdk.')).

    Follow the advice of the message and run python -m pip install --verbose --force-reinstall oneagent-sdk (or the equivalent pip invocation with the --verbose and --force-reinstall flags). It is likely that you will now see another message like

      *** You are trying to build the Python SDK from source.                    ***
      *** This could mean that you are using an outdated version of pip (older   ***
      *** than 8.1.0) or you are attempting to install the SDK on an             ***
      *** unsupported platform. Please check the requirements at                 ***
      ***      ***

    Make sure you are using pip to install a prebuilt package wheel for your system from PyPI, as described in Using the OneAgent SDK for Python in your application. Also make sure you are using an up-to date version of pip, setuptools and wheel. You can try upgrading them with python -m pip install --upgrade pip setuptools wheel (make sure to use the same python that you use to install the oneagent-sdk package). ATTENTION: If you use the system-provided pip (e.g. installed via apt-get on Ubuntu) you should instead use a pip inside a virtualenv (the same as your project), as upgrading system-provided packages via pip may cause issues.

    If this does not resolve the issue, make sure you are using a supported platform, as listed in Requirements. If you are using a supported system, you can try downloading the OneAgent SDK for C/C++ in the version corresponding to your OneAgent SDK for Python as listed in the table in Requirements. Then set the DT_PYSDK_CSDK_PATH environment variable to the .so/.dll file corresponding to your platform in the lib subdirectory of the C SDK and retry the installation (e.g. in a bash shell, use export DT_PYSDK_CSDK_PATH=path/to/ If there is no corresponding directory, your platform is not supported. Otherwise, regardless if it works with that method or not, please report an issue as described in Let us help you.

Extended SDK State

For debugging and/or diagnosing purposes you can also use the extended SDK state information.

# The agent state is one of the integers in oneagent.sdk.AgentState.
print('Agent state:', oneagent.get_sdk().agent_state)

# The instance attribute 'agent_found' indicates whether an agent could be found or not.
print('Agent found:', oneagent.get_sdk().agent_found)

# If an agent was found but it is incompatible with this version of the SDK for Python
# then 'agent_is_compatible' would be set to false.
print('Agent is compatible:', oneagent.get_sdk().agent_is_compatible)

# The agent version is a string holding both the OneAgent version and the
# OneAgent SDK for C/C++ version separated by a '/'.
print('Agent version:', oneagent.get_sdk().agent_version_string)

Shutdown crashes

If your are experiencing crashes when your application exits, make sure the you uninitialized the SDK properly by calling its shutdown function.

Repository contents

If you are viewing the GitHub repository, you will see:

  • LICENSE: License under which the whole SDK and sample applications are published.
  • src/: Actual source code of the Python OneAgent SDK.
  • docs/: Source files for the (Sphinx-based) HTML documentation. For the actual, readable documentation, see here.
  • tests/, test-util-src/: Contains tests and test support files that are useful (only) for developers wanting to contribute to the SDK itself.
  •, setup.cfg,, project.toml: Development files required for creating e.g. the PyPI package for the Python OneAgent SDK.
  • tox.ini, pylintrc: Supporting files for developing the SDK itself. See and

Help & Support

Support policy

The Dynatrace OneAgent SDK for Python has GA status. The features are fully supported by Dynatrace.

For detailed support policy see Dynatrace OneAgent SDK help.

Read the manual

Let us help you

Make sure your issue is not already solved in the available documentation before you ask for help. Especially the troubleshooting section in this README may prove helpful.

Get Help

Open a GitHub issue to:

  • Report minor defects or typos.
  • Ask for improvements or changes in the SDK API.
  • Ask any questions related to the community effort.

SLAs don't apply for GitHub tickets.

Customers can open a ticket on the Dynatrace support portal to:

  • Get support from the Dynatrace technical support engineering team
  • Manage and resolve product related technical issues

SLAs apply according to the customer's support level.

Release notes

Version 1.4.0

  • Don't look for agent module in PATH/LD_LIBRARY_PATH/... and disallow relative a DT_HOME directory on Windows to prevent DLL hijacking issues.

  • Fixed a bug that might lead to crashes in the SDK's shutdown phase

  • Support for Python versions < 3.5 is deprecated. The OneAgent SDK for Python will still work with this release, but this might change in the future.

  • Following versions of the OneAgent SDK for Python are considered deprecated and might not be supported in the future. Applications using it should be upgraded to the latest release.

For older versions of the OneAgent SDK for Python, please see the GitHub releases page, and the PyPI release history.


See the LICENSE file for details. It should be included in your distribution. Otherwise, see the most recent version on GitHub.

Summary: This software is licensed under the terms of the Apache License Version 2.0 and comes bundled with the six library by Benjamin Peterson, which is licensed under the terms of the MIT license.

Project details

Download files

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

Source Distribution (89.4 kB view hashes)

Uploaded source

Built Distributions

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page