AWS4 authentication for Requests
Amazon Web Services version 4 authentication for the Python Requests library.
Implements header-based authentication, GET URL parameter and POST parameter authentication are not supported.
This package has been tested as working against:
AppStream, Auto-Scaling, CloudFormation, CloudFront, CloudHSM, CloudSearch, CloudTrail, CloudWatch Monitoring, CloudWatch Logs, CodeDeploy, Cognito Identity, Cognito Sync, Config, DataPipeline, Direct Connect, DynamoDB, Elastic Beanstalk, ElastiCache, EC2, EC2 Container Service, Elastic Load Balancing, Elastic MapReduce, ElasticSearch, Elastic Transcoder, Glacier, Identity and Access Management (IAM), Key Management Service (KMS), Kinesis, Lambda, Opsworks, Redshift, Relational Database Service (RDS), Route 53, Simple Storage Service (S3), Simple Notification Service (SNS), Simple Queue Service (SQS), Storage Gateway, Security Token Service (STS)
The following services do not support AWS auth version 4 and are not usable with this package:
Simple Email Service (SES), Simple Workflow Service (SWF), Import/Export, SimpleDB, DevPay, Mechanical Turk
The AWS Support API has not been tested as it requires a premium subscription.
Install via pip:
$ pip install requests-aws4auth
requests-aws4auth requires the Requests library by Kenneth Reitz.
requests-aws4auth is tested on Python 2.7 and 3.3 and up.
Version 0.8 introduces request date checking and automatic key regeneration behaviour as default. This has implications for sharing authentication objects between threads, and for storage of secret keys. See the relevant sections below for details. See also the discussion in GitHub issue #10.
>>> import requests >>> from requests_aws4auth import AWS4Auth >>> endpoint = 'http://s3-eu-west-1.amazonaws.com' >>> auth = AWS4Auth('<ACCESS ID>', '<ACCESS KEY>', 'eu-west-1', 's3') >>> response = requests.get(endpoint, auth=auth) >>> response.text <?xml version="1.0" encoding="UTF-8"?> <ListAllMyBucketsResult xmlns="http://s3.amazonaws.com/doc/2006-03-01"> <Owner> <ID>bcaf1ffd86f461ca5fb16fd081034f</ID> <DisplayName>webfile</DisplayName> ...
This example would list your buckets in the eu-west-1 region of the Amazon S3 service.
>>> from requests_aws4auth import AWS4Auth >>> auth = AWS4Auth('<ACCESS ID>', '<ACCESS KEY>', 'eu-west-1', 's3', session_token='<SESSION TOKEN>') ...
This example shows how to construct an AWS4Auth object for use with STS temporary credentials. The x-amz-security-token header is added with the session token. Temporary credential timeouts are not managed – in case the temporary credentials expire, they need to be re-generated and the AWS4Auth object re-constructed with the new credentials.
If an HTTP request to be authenticated contains a Date or X-Amz-Date header, AWS will only accept the authorised request if the date in the header matches the scope date of the signing key (see the AWS REST API date docs).
From version 0.8 of requests-aws4auth, if the header date does not match the scope date, an AWS4Auth instance will automatically regenerate its signing key, using the same scope parameters as the previous key except for the date, which will be changed to match the request date. If a request does not include a date, the current date is added to the request in an X-Amz-Date header, and the signing key is regenerated if this differs from the scope date.
This means that AWS4Auth now extracts and parses dates from the values of X-Amz-Date and Date headers. Supported date formats are:
If either header is present but AWS4Auth cannot extract a date because all present date headers are in an unrecognisable format, AWS4Auth will delete any X-Amz-Date and Date headers present and replace with a single X-Amz-Date header containing the current date. This behaviour can be modified using the raise_invalid_date keyword argument of the AWS4Auth constructor.
If you do not want the signing key to be automatically regenerated when a mismatch between the request date and the scope date is encountered, use the alternative StrictAWS4Auth class, which is identical to AWS4Auth except that upon encountering a date mismatch it just raises a DateMismatchError. You can also use the PassiveAWS4Auth class, which mimics the AWS4Auth behaviour prior to version 0.8 and just signs and sends the request, whether the date matches or not. In this case it is up to the calling code to handle an authentication failure response from AWS caused by the date mismatch.
To allow automatic key regeneration, the secret key is stored in the AWS4Auth instance, in the signing key object. If you do not want this to occur, instantiate the instance using an AWS4Signing key which was created with the store_secret_key parameter set to False:
>>> sig_key = AWS4SigningKey(secret_key, region, service, date, False) >>> auth = StrictAWS4Auth(access_id, sig_key)
The AWS4Auth class will then raise a NoSecretKeyError when it attempts to regenerate its key. A slightly more conceptually elegant way to handle this is to use the alternative StrictAWS4Auth class, again instantiating it with an AWS4SigningKey instance created with store_secret_key = False.
If you share AWS4Auth (or even StrictAWS4Auth) instances between threads you are likely to encounter problems. Because AWS4Auth instances may unpredictably regenerate their signing key as part of signing a request, threads using the same instance may find the key changed by another thread halfway through the signing process, which may result in undefined behaviour.
It may be possible to rig up a workable instance sharing mechanism using locking primitives and the StrictAWS4Auth class, however this poor author can’t think of a scenario which works safely yet doesn’t suffer from at some point blocking all threads for at least the duration of an HTTP request, which could be several seconds. If several requests come in in close succession which all require key regenerations then the system could be forced into serial operation for quite a length of time.
In short, it’s probably best to create a thread-local instance of AWS4Auth for each thread that needs to do authentication.
See the doctrings in aws4auth.py and aws4signingkey.py.
A test suite is included in the test folder.
The package passes all tests in the AWS auth v4 test_suite, and contains tests against the supported live services. See docstrings in test/requests_aws4auth_test.py for details about running the tests.
Connection parameters are included in the tests for the AWS Support API, should you have access and want to try it. The documentation says it supports auth v4 so it should work if you have a subscription. Do pass on your results!
This version introduces some behaviour changes designed to reduce the legwork needed when a signing key goes out of date. This has implications for multithreading and secret key storage. See the README for further details.