Changes in Apache Libcloud v2.0¶
Replacement of httplib with requests¶
Apache Libcloud supports Python 2.6, 2.7 - 3.3 and beyond. To achieve this a package was written within the Libcloud library to create a generic HTTP client for Python 2 and 3. This package has a custom implementation of a certificate store, searching and TLS preference configuration. One of the first errors to greet new users of Libcloud would be “No CA Certificates were found in CA_CERTS_PATH.”…
In 2.0 this implementation has been replaced with the requests package, and SSL verification should work against any publicly signed HTTPS endpoint by default, without having to provide a CA cert store.
Other changes include:
Enabling HTTP redirects
Allowing both global and driver-specific HTTP proxy configuration
Consolidation of the LibcloudHTTPSConnection and LibcloudHTTPConnection into a single class, LibcloudConnection
Support for streaming responses
Support for mocking HTTP responses without having to mock the Connection class
10% typical performance improvement with the use of persistent TCP connections for each driver instance
Access to the low-level TCP session is no longer available. Access to .read() on a raw connection will bind around requests body or iter_content methods.
Temporary removal of the S3 very-large file support using the custom multi-part APIs. This will be added back in subsequent release candidates.
Allow redirects is enabled by default¶
HTTP redirects are allowed by default in 2.0. To disable redirects, set this global variable to False.
import libcloud.http libcloud.http.ALLOW_REDIRECTS = False
Enabling a HTTP/HTTPS proxy is still supported and accessed via the driver’s connection property or via the ‘http_proxy’ environment variable. Applying it to a driver will set the proxy for that driver only, using the environment variable will make a global change.
# option 1 import os os.environ.get('http_proxy', 'http://localhost:8888/') # option 2 driver.connection.connection.set_http_proxy(proxy_url='http://localhost:8888')
Adding support for Python 3.6 and deprecation of Python 3.2¶
In Apache Libcloud 2.0.0, Python 3.6 is now supported as a primary distribution.
Python 3.2 support has been dropped in this release and users should either upgrade to 3.3 or a newer version of Python.
SSL CA certificates are now bundled with the package¶
In Apache Libcloud 2.0.0, the Mozilla Trusted Root Store is bundled with the package, as part of the requests package bundle. This means that users no longer have to set the path to a CA file either via installing the certifi package, downloading a PEM file or providing a directory in an environment variable. All connections in Libcloud will assume HTTPS by default, now with 2.0.0, if those HTTPS endpoints have a signed certificate with a trusted CA authority, they will work with Libcloud by default.
Providing a custom client-side certificate, for example for a development server or a HTTPS proxy is still supported given providing a value to libcloud.security.CA_CERTS_PATH.
This code example would set a HTTP/HTTPS proxy and use a client-generated certificate to verify.
import os os.environ.set('http_proxy', 'http://localhost:8888/') import libcloud.security libcloud.security.VERIFY_SSL_CERT = True libcloud.security.CA_CERTS_PATH = '/Users/anthonyshaw/charles.pem'
Providing a list of CA trusts is no longer supported¶
In Apache Libcloud 2.0.0 if you provide a list of more than 1 path or certificate file in libcloud.security.CA_CERTS_PATH you will receive a warning and only the first path will be used. This path should be to a .cert or .pem file. The environment variable REQUESTS_CA_BUNDLE can be used to access the requests library’s list of trusted CAs.
Performance improvements and introduction of sessions¶
Each instance of libcloud.common.base.Connection will have a LibcloudConnection instance under the connection property. In 1.5.0<, there would be 2 connection
class instances, LibcloudHttpConnection and LibcloudHttpsConnection, stored as an instance property conn_classes. In 2.0.0 this has been replaced with a single type,
libcloud.common.base.LibcloudHTTPConnection that handles both HTTP and HTTPS connections.
def test(): import libcloud import libcloud.compute.providers d = libcloud.get_driver(libcloud.DriverType.COMPUTE, libcloud.DriverType.COMPUTE.DIMENSIONDATA) instance = d('anthony', 'mypassword!', 'dd-au') instance.list_nodes() # is paged instance.list_images() # is paged if __name__ == '__main__': import timeit print(timeit.timeit("test()", setup="from __main__ import test", number=5))
This simple test shows a 10% performance improvement between Libcloud 1.5.0 and 2.0.0.
Changes to the storage API¶
Support for Buffered IO Streams¶
The methods upload_object_via_stream now supports file objects, BytesIO, StringIO and generators as the iterator.
with open('my_file_to_upload', 'rb') as iterator: obj = driver.upload_object_via_stream(iterator=iterator, container=containers, object_name='me.jpg', extra=extra)
Other minor changes¶
libcloud.common.base.Connectionwill now use urljoin to combine the request_path and method URLs. This means that the URL action will always have a leading slash.
The underlying connection classes do not assume HTTP if a non-standard port is used. They will use the preference set in the secure flag to the initializer of Connection.
The storage download_object_as_stream method no longer buffers out file streams twice.