Skip to main content

Microsoft Azure Core Library for Python

Project description

Azure Core Library


The Azure Core pipeline is a re-structuring of the msrest pipeline introduced in msrest 0.6.0. Further discussions on the msrest implementation can be found in the msrest wiki.

The Azure Core Pipeline is an implementation of chained policies as described in the Azure SDK guidelines.

The Python implementation of the pipeline has some mechanisms specific to Python. This is due to the fact that both synchronous and asynchronous implementations of the pipeline must be supported independently.

When constructing an SDK, a developer may consume the pipeline like so:

from azure.core import Configuration, Pipeline
from azure.core.transport import RequestsTransport, HttpRequest
from azure.core.pipeline.policies import (

class FooServiceClient:

    def _create_config(credential, scopes, **kwargs):
        # Here the SDK developer would define the default
        # config to interact with the service
        config = Configuration(**kwargs)
        config.headers_policy = kwargs.get('headers_policy', HeadersPolicy({"CustomHeader": "Value"}, **kwargs))
        config.user_agent_policy = kwargs.get('user_agent_policy', UserAgentPolicy("ServiceUserAgentValue", **kwargs))
        config.authentication_policy = kwargs.get('authentication_policy', BearerTokenCredentialPolicy(credential, scopes, **kwargs))
        config.retry_policy = kwargs.get('retry_policy', RetryPolicy(**kwargs))
        config.redirect_policy = kwargs.get('redirect_policy', RedirectPolicy(**kwargs))
        config.logging_policy = kwargs.get('logging_policy', NetworkTraceLoggingPolicy(**kwargs))
        config.proxy_policy = kwargs.get('proxy_policy', ProxyPolicy(**kwargs))
        return config

    def __init__(self, **kwargs):
        transport = kwargs.get('transport', RequestsTransport(**kwargs))
        config = FooServiceClient._create_config(**kwargs)
        policies = [
        self._pipeline = Pipeline(transport, policies=policies)

    def get_foo_properties(self, **kwargs)
        # Create a generic HTTP Request. This is not specific to any particular transport
        # or pipeline configuration.
        new_request = HttpRequest("GET", "/")

        response =, **kwargs)
        return deserialize_data(response.http_response)

An end user consuming this SDK may write code like so:

from azure.core.credentials import FooCredentials
from import FooServiceClient

creds = FooCredentials("api-key")
endpoint = "

# Scenario using entirely default configuration
# We use the SDK-developer defined configuration.
client = FooServiceClient(endpoint, creds)
response = client.get_foo_properties()

# Scenario where user wishes to tweak a couple of settings
# In this case the configurable options can be passed directly into the client constructor.
client = FooServiceClient(endpoint, creds, logging_enable=True, retries_total=5)
response = client.get_foo_properties()

# Scenario where user wishes to tweak settings for only a specific request
# All the options available on construction are available as per-request overrides.
# These can also be specified by the SDK developer - and it will be up to them to resolve
# conflicts with user-defined parameters.
client = FooServiceClient(endpoint, creds)
response = client.get_foo_properties(redirects_max=0)

# Scenario where user wishes to fully customize the policies.
# All configuration options are passed through kwargs
client = FooServiceClient(
response = client.get_foo_properties()


The Configuration object is the home of all the configurable policies in the pipeline. A new Configuration object provides no default policies. It is up to the SDK developer to specify each of the policy defaults as required by the service.

Configuration should not be exposed as part of the public API of the resulting SDK.

This can be seen in the above code sample as implemented in a staticmethod on the client class. The Configuration object does not specify in what order the policies will be added to the pipeline. It is up to the SDK developer to use the policies in the Configuration to construct the pipeline correctly, as well as inserting any unexposed/non-configurable policies.

transport = RequestsTransport(**kwargs)

# SDK developer needs to build the policy order for the pipeline.
config = FooServiceClient._create_config(**kwargs)
policies = [
    config.authentication_policy,  # Authentication policy needs to be inserted after all request mutation to accommodate signing.
    config.logging_policy,  # Logger should come last to accurately record the request/response as they are on the wire
self._pipeline = Pipeline(transport, policies=policies)

The policies that should currently be defined on the Configuration object are as follows:

- Configuration.headers_policy  # HeadersPolicy
- Configuration.retry_policy  # RetryPolicy
- Configuration.redirect_policy  # RedirectPolicy
- Configuration.logging_policy  # NetworkTraceLoggingPolicy
- Configuration.user_agent_policy  # UserAgentPolicy
- Configuration.proxy_policy  # While this is a ProxyPolicy object, current implementation is transport configuration.
- Configuration.authentication_policy  # BearerTokenCredentialPolicy


Various combinations of sync/async HTTP libraries as well as alternative event loop implementations are available. Therefore to support the widest range of customer scenarios, we must allow a customer to easily swap out the HTTP transport layer to one of those supported.

The transport is the last node in the pipeline, and adheres to the same basic API as any policy within the pipeline. The only currently available transport for synchronous pipelines uses the Requests library:

from azure.core.pipeline.transport import RequestsTransport
synchronous_transport = RequestsTransport()

For asynchronous pipelines a couple of transport options are available. Each of these transports are interchangable depending on whether the user has installed various 3rd party dependencies (i.e. aiohttp or trio), and the user should easily be able to specify their chosen transport. SDK developers should use the aiohttp transport as the default for asynchronous pipelines where the user has not specified an alternative.

from import FooServiceClient
from azure.core.pipeline.transport import (
    # Identical implementation as the synchronous RequestsTransport wrapped in an asynchronous using the
    # built-in asyncio event loop.

    # Identical implementation as the synchronous RequestsTransport wrapped in an asynchronous using the
    # third party trio event loop.

    # Fully asynchronous implementation using the aiohttp library, using the built-in asyncio event loop.

client = FooServiceClient(endpoint, creds, transport=AioHttpTransport())
response = await client.get_foo_properties()

Some common properties can be configured on all transports. They must be passed as kwargs arguments while building the transport instance. These include the following properties:

transport = AioHttpTransport(
        # The connect and read timeout value. Defaults to 100 seconds.

        # SSL certificate verification. Enabled by default. Set to False to disable,
        # alternatively can be set to the path to a CA_BUNDLE file or directory with
        # certificates of trusted CAs.

        # Client-side certificates. You can specify a local cert to use as client side
        # certificate, as a single file (containing the private key and the certificate)
        # or as a # tuple of both files' paths.

        # The block size of data sent over the connection. Defaults to 4096 bytes.

HttpRequest and HttpResponse

The HttpRequest and HttpResponse objects represent a generic concept of HTTP request and response constructs and are in no way tied to a particular transport or HTTP library.

The HttpRequest has the following API. It does not vary between transports:

class HttpRequest(object):

    def __init__(self, method, url, headers=None, files=None, data=None):
        self.method = method
        self.url = url
        self.headers = CaseInsensitiveDict(headers)
        self.files = files = data

    def body(self):

    def body(self, value): = value

    def format_parameters(self, params):
        """Format parameters into a valid query string.
        It's assumed all parameters have already been quoted as
        valid URL strings."""

    def set_xml_body(self, data):
        """Set an XML element tree as the body of the request."""

    def set_json_body(self, data):
        """Set a JSON-friendly object as the body of the request."""

    def set_multipart_body(self, data=None):
        """Set form-encoded data as the body of the request.
        Supported content-types are:
            - application/x-www-form-urlencoded
            - multipart/form-data

    def set_bytes_body(self, data):
        """Set generic bytes as the body of the request."""

The HttpResponse object on the other hand will generally have a transport-specific derivative. This is to accomodate how the data is extracted for the object returned by the HTTP library. There is also an async flavor: AsyncHttpResponse. This is to allow for the asynchronous streaming of data from the response. For example:

from azure.core.pipeline.transport import (
    RequestsTransportResponse,  # HttpResponse
    AioHttpTransportResponse, # AsyncHttpResponse
    TrioRequestsTransportResponse,  # AsyncHttpResponse
    AsyncioRequestsTransportResponse,  # AsyncHttpResponse

The API for each of these response types is identical, so the consumer of the Response need not know about these particular types.

The HttpResponse has the following API. It does not vary between transports:

class HttpResponse(object):

    def __init__(self, request, internal_response):
        self.request = request
        self.internal_response = internal_response  # The object returned by the HTTP library
        self.status_code = None
        self.headers = CaseInsensitiveDict()
        self.reason = None
        self.content_type = None

    def body(self):
        """Return the whole body as bytes in memory."""

    def text(self, encoding=None):
        """Return the whole body as a string."""

    def stream_download(self, chunk_size=None, callback=None):
        """Generator for streaming request body data.
        Should be implemented by sub-classes if streaming download
        is supported.
        For the AsyncHttpResponse object this function will return
        and asynchronous generator.

PipelineRequest and PipelineResponse

These objects provide containers for moving the HttpRequest and HttpResponse through the pipeline. While the SDK developer will not construct the PipelineRequest explicitly, they will handle the PipelineResponse object that is returned from These objects are universal for all transports, both synchronous and asynchronous.

The pipeline request and response containers are also responsible for carrying a context object. This is transport specific and can contain data persisted between pipeline requests (for example reusing an open connection pool or "session"), as well as used by the SDK developer to carry arbitrary data through the pipeline.

The API for PipelineRequest and PipelineResponse is as follows:

class PipelineRequest(object):

    def __init__(self, http_request, context):
        self.http_request = http_request  # The HttpRequest
        self.context = context # A transport specific data container object

class PipelineResponse(object):

    def __init__(self, http_request, http_response, context):
        self.http_request = http_request  # The HttpRequest
        self.http_response = http_response  # The HttpResponse
        self.history = []  # A list of redirect attempts.
        self.context = context  # A transport specific data container object


The Python pipeline implementation provides two flavors of policy. These are referred to as an HttpPolicy and a SansIOHTTPPolicy.


If a policy just modifies or annotates the request based on the HTTP specification, it's then a subclass of SansIOHTTPPolicy and will work in either Pipeline or AsyncPipeline context. This is a simple abstract class, that can act before the request is done, or after. For instance:

  • Setting headers in the request
  • Logging the request and/or response

A SansIOHTTPPolicy should implement one or more of the following methods:

def on_request(self, request):
    """Is executed before sending the request to next policy."""

def on_response(self, request, response):
    """Is executed after the request comes back from the policy."""

def on_exception(self, request):
    """Is executed if an exception is raised while executing this policy.

    Return True if the exception has been handled and should not
    be forwarded to the caller.

Current provided sans IO policies include:

from azure.core.pipeline.policies import (
    HeadersPolicy,  # Add custom headers to all requests
    UserAgentPolicy,  # Add a custom user agent header
    NetworkTraceLoggingPolicy,  # Log request and response contents
    ContentDecodePolicy,  # Mandatory policy for decoding unstreamed response content

HTTPPolicy and AsyncHTTPPolicy

Some policies are more complex, like retry strategy, and need to have control of the HTTP workflow. In the current version, they are subclasses of HTTPPolicy or AsyncHTTPPolicy, and can be used only their corresponding synchronous or asynchronous pipeline type.

An HTTPPolicy or AsyncHTTPPolicy must implement the send method, and this implementation must in include a call to process the next policy in the pipeline:

class CustomPolicy(HTTPPolicy):

    def __init__(self): = None  # Will be set when pipeline is instantiated and all the policies chained.

    def send(self, request):
        """Mutate the request."""


class CustomAsyncPolicy(AsyncHTTPPolicy):

    async def send(self, request):
        """Mutate the request."""

        return await

Currently provided HTTP policies include:

from azure.core.pipeline.policies import (

The Pipeline

The pipeline itself represents a chain of policies where the final node in the chain is the HTTP transport. A pipeline can either be synchronous or asynchronous. The pipeline does not expose the policy chain, so individual policies cannot/should not be further configured once the pipeline has been instantiated.

The pipeline has a single exposed operation: run(request) which will send a new HttpRequest object down the pipeline. This operation returns a PipelineResponse object.

class Pipeline:
    """A pipeline implementation.

    This is implemented as a context manager, that will activate the context
    of the HTTP sender.

    def __init__(self, transport, policies=None):
        # type: (HttpTransport, List[Union[HTTPPolicy, SansIOHTTPPolicy]]) -> None
        self._impl_policies = []  # type: List[HTTPPolicy]
        self._transport = transport  # type: HTTPPolicy

        for policy in (policies or []):
            if isinstance(policy, SansIOHTTPPolicy):
            elif policy:
        for index in range(len(self._impl_policies)-1):
            self._impl_policies[index].next = self._impl_policies[index+1]
        if self._impl_policies:
            self._impl_policies[-1].next = _TransportRunner(self._transport)

    def run(self, request, **kwargs):
        # type: (HTTPRequestType, Any) -> PipelineResponse
        context = PipelineContext(self._transport, **kwargs)
        pipeline_request = PipelineRequest(request, context)  # type: PipelineRequest[HTTPRequestType]
        first_node = self._impl_policies[0] if self._impl_policies else _TransportRunner(self._transport)
        return first_node.send(pipeline_request)  # type: ignore

Release History

2019-08-XX Version 1.0.0b2

Breaking changes

  • Transport classes don't take config parameter anymore (use kwargs instead) #6372
  • azure.core.paging has been completely refactored #6420
  • HttpResponse.content_type attribute is now a string (was a list) #6490
  • For StreamDownloadGenerator subclasses, response is now an HttpResponse, and not a transport response like aiohttp.ClientResponse or requests.Response. The transport response is available in internal_response attribute #6490

Bug fixes

  • aiohttp is not required to import async pipelines classes #6496
  • AsyncioRequestsTransport.sleep is now a coroutine as expected #6490
  • RequestsTransport is not tight to ProxyPolicy implementation details anymore #6372
  • AiohttpTransport does not raise on unexpected kwargs #6355


  • New paging base classes that support continuation_token and by_page() #6420
  • Proxy support for AiohttpTransport #6372

2019-06-26 Version 1.0.0b1

  • Preview 1 release

Project details

Download files

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

Files for azure-core, version 1.0.0b2
Filename, size File type Python version Upload date Hashes
Filename, size azure_core-1.0.0b2-py2.py3-none-any.whl (98.3 kB) File type Wheel Python version py2.py3 Upload date Hashes View
Filename, size (144.3 kB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page