Rule Development Kit - Version 2 This tool should be considered in "Open Beta". I would greatly appreciate feedback and bug reports at [email protected]!
The RDK is designed to support a "Compliance-as-Code" workflow that is intuitive and productive. It abstracts away much of the undifferentiated heavy lifting associated with deploying AWS Config rules backed by custom lambda functions, and provides a streamlined develop-deploy-monitor iterative process.
Uses python 2.7/3.6 and is installed via pip. Requires you to have an AWS account and sufficient permissions to manage the Config service, and to create S3 Buckets, Roles, and Lambda Functions. An AWS IAM Policy Document that describes the minimum necessary permissions can be found at policy/rdk-minimum-permissions.json.
Under the hood, rdk uses boto3 to make API calls to AWS, so you can set your credentials any way that boto3 recognizes (options 3 through 8 here: http://boto3.readthedocs.io/en/latest/guide/configuration.html) or pass them in with the command-line parameters --profile, --region, --access-key-id, or --secret-access-key
If you just want to use the RDK, go ahead and install it using pip:
$ pip install rdk
Alternately, if you want to see the code and/or contribute you can clone the git repo, and then from the repo directory use pip to install the package. Use the '-e' flag to generate symlinks so that any edits you make will be reflected when you run the installed package.
If you are going to author your Lambda functions using Java you will need to have Java 8 and gradle installed. If you are going to author your Lambda functions in C# you will need to have the dotnet CLI and the .NET Core Runtime 1.08 installed.
$ pip install -e .
To make sure the rdk is installed correctly, running the package from the command line without any arguments should display help information.
$ rdk
usage: rdk [-h] [-p PROFILE] [-k ACCESS_KEY] [-s SECRET_ACCESS_KEY]
[-r REGION]
<command> ...
rdk: error: the following arguments are required: <command>, <command arguments>
To use the RDK, it's recommended to create a directory that will be your working directory. This should be committed to a source code repo, and ideally created as a python virtualenv. In that directory, run the init command to set up your AWS Config environment.
$ rdk init Running init! Creating Config bucket config-bucket-780784666283 Creating IAM role config-role Waiting for IAM role to propagate Config Service is ON Config setup complete. Creating Code bucket config-rule-code-bucket-780784666283ap-southeast-1
Running init subsequent times will validate your AWS Config setup and re-create any S3 buckets or IAM resources that are needed.
In your working directory, use the create command to start creating a new custom rule. You must specify the runtime for the lambda function that will back the Rule, and you can also specify a resource type (or comma-separated list of types) that the Rule will evaluate or a maximum frequency for a periodic rule. This will add a new directory for the rule and populate it with several files, including a skeleton of your Lambda code.
$ rdk create MyRule --runtime python3.6 --resource-types AWS::EC2::Instance --input-parameters '{"desiredInstanceType":"t2.micro"}'
Running create!
Local Rule files created.
On Windows it is necessary to escape the double-quotes when specifying input parameters, so the --input-parameters argument would instead look something like this:
'{\"desiredInstanceType\":\"t2.micro\"}'
Note that you can create rules that use EITHER resource-types OR maximum-frequency, but not both. We have found that rules that try to be both event-triggered as well as periodic wind up being very complicated and so we do not recommend it as a best practice.
Once you have created the rule, edit the python file in your rule directory (in the above example it would be MyRule/MyRule.py, but may be deeper into the rule directory tree depending on your chosen Lambda runtime) to add whatever logic your Rule requires in the evaluate_compliance function. You will have access to the CI that was sent by Config, as well as any parameters configured for the Config Rule. Your function should return either a simple compliance status (one of COMPLIANT, NONCOMPLIANT, or NOT_APPLICABLE), or if you're using the python or node runtimes you can return a JSON object with multiple evaluation responses that the RDK will send back to AWS Config. An example would look like:
for sg in response['SecurityGroups']:
evaluations.append(
{
'ComplianceResourceType': 'AWS::EC2::SecurityGroup',
'ComplianceResourceId': sg['GroupId'],
'ComplianceType': 'COMPLIANT',
'Annotation': 'This is an important note.',
'OrderingTimestamp': str(datetime.datetime.now())
})
return evaluations
This is necessary for periodic rules that are not triggered by any CI change (which means the CI that is passed in will be null), and also for attaching annotations to your evaluation results.
If you want to see what the JSON structure of a CI looks like for creating your logic, you can use
$ rdk sample-ci <Resource Type>
to output a formatted JSON document.
If you are writing Config Rules using either of the Python runtimes there will be a <rule name>_test.py file deployed along with your Lambda function skeleton. This can be used to write unit tests according to the standard Python unittest framework (documented here: https://docs.python.org/3/library/unittest.html), which can be run using the test-local rdk command:
$ rdk test-local MyTestRule Running local test! Testing MyTestRule Looking for tests in /Users/mborch/Code/rdk-dev/MyTestRule --------------------------------------------------------------------- Ran 0 tests in 0.000s OK <unittest.runner.TextTestResult run=0 errors=0 failures=0>
The test file includes setup for the MagicMock library that can be used to stub boto3 API calls if your rule logic will involve making API calls to gather additional information about your AWS environment. For some tips on how to do this, check out this blog post: https://sgillies.net/2017/10/19/mock-is-magic.html
If you need to change the parameters of a Config rule in your working directory you can use the modify command. Any parameters you specify will overwrite existing values, any that you do not specify will not be changed.
$ rdk modify MyRule --runtime python2.7 --maximum-frequency TwentyFour_Hours --input-parameters '{"desiredInstanceType":"t2.micro"}'
Running modify!
Modified Rule 'MyRule'. Use the `deploy` command to push your changes to AWS.
Again, on Windows the input parameters would look like:
'{\"desiredInstanceType\":\"t2.micro\"}'
It is worth noting that until you actually call the deploy command your rule only exists in your working directory, none of the Rule commands discussed thus far actually makes changes to your account.
Once you have completed your compliance validation code and set your Rule's configuration, you can deploy the Rule to your account using the deploy command. This will zip up your code (and the other associated code files, if any) into a deployable package (or run a gradle build if you have selected the java8 runtime or run the lambda packaging step from the dotnet CLI if you have selected the dotnetcore1.0 runtime), copy that zip file to S3, and then launch or update a CloudFormation stack that defines your Config Rule, Lambda function, and the necessary permissions and IAM Roles for it to function. Since CloudFormation does not deeply inspect Lambda code objects in S3 to construct its changeset, the deploy command will also directly update the Lambda function for any subsequent deployments to make sure code changes are propagated correctly.
$ rdk deploy MyRule Running deploy! Zipping MyRule Uploading MyRule Creating CloudFormation Stack for MyRule Waiting for CloudFormation stack operation to complete... ... Waiting for CloudFormation stack operation to complete... Config deploy complete.
The exact output will vary depending on Lambda runtime. You can use the --all flag to deploy all of the rules in your working directory.
Once the Rule has been deployed to AWS you can get the CloudWatch logs associated with your lambda function using the logs command.
$ rdk logs MyRule -n 5
2017-11-15 22:59:33 - START RequestId: 96e7639a-ca15-11e7-95a2-b1521890638d Version: $LATEST
2017-11-15 23:41:13 - REPORT RequestId: 68e0304f-ca1b-11e7-b735-81ebae95acda Duration: 0.50 ms Billed Duration: 100 ms Memory Size: 256 MB
Max Memory Used: 36 MB
2017-11-15 23:41:13 - END RequestId: 68e0304f-ca1b-11e7-b735-81ebae95acda
2017-11-15 23:41:13 - Default RDK utility class does not yet support Scheduled Notifications.
2017-11-15 23:41:13 - START RequestId: 68e0304f-ca1b-11e7-b735-81ebae95acda Version: $LATEST
You can use the -n and -f command line flags just like the UNIX tail command to view a larger number of log events and to continuously poll for new events. The latter option can be useful in conjunction with manually initiating Config Evaluations for your deploy Config Rule to make sure it is behaving as expected.
The testing directory contains scripts and buildspec files that I use to run basic functionality tests across a variety of CLI environemnts (currently Ubuntu linux running python2.7, Ubuntu linux running python 3.6, and Windows Server running python2.7). If there is interest I can release a CloudFormation template that could be used to build the test environment, let me know if this is something you want!
Features have been added to the RDK to facilitate the cross-account deployment pattern that enterprise customers have standardized on for custom Config Rules. A cross-account architecutre is one in which the Lambda functions are deployed to a single central "Compliance" account (which may be the same as a central "Security" account), and the Config Rules are deployed to any number of "Satellite" accounts that are used by other teams or departments. This gives the compliance team confidence that their Rule logic cannot be tampered with and makes it much easier for them to modify rule logic without having to go through a complex deployment process to potentially hundreds of AWS accounts. The cross-account pattern uses two advanced RDK features - functions-only deployments and the create-rule-template command.
Function-Only Deployment
By using the -f or --functions-only flag on the deploy command the RDK will deploy only the necessary Lambda Functions, Lambda Execution Role, and Lambda Permissions to the account specified by the execution credentials. It accomplishes this by batching up all of the Lambda function CloudFormation snippets for the selected Rule(s) into a single dynamically generated template and deploy that CloudFormation template. One consequence of this is that subsequent deployments that specify a different set of Rules for the same stack name will update that CloudFormation stack, and any Rules that were included in the first deployment but not in the second will be removed. You can use the --stack-name parameter to override the default CloudFormation stack name if you need to manage different subsets of your Lambda Functions independently. The intended usage is to deploy the functions for all of the Config rules in the Security/Compliance account, which can be done simply by using rdk deploy -f --all from your working directory.
`create-rule-template` command
This command generates a CloudFormation template that defines the AWS Config rules themselves, along with the Config Role, Config data bucket, Configuration Recorder, and Delivery channel necessary for the Config rules to work in a satellite account. You must specify the file name for the generated template using the --output-file or o command line flags. The generated template takes a single parameter of the AccountID of the central compliance account that contains the Lambda functions that will back your custom Config Rules. The generated template can be deployed in the desired satellite accounts through any of the means that you can deploy any other CloudFormation template, including the console, the CLI, as a CodePipeline task, or using StackSets. The create-rule-template command takes all of the standard arguments for selecting Rules to include in the generated template, including lists of individual Rule names, an --all flag, or using the RuleSets feature described below.
$ rdk create-rule-template -o remote-rule-template.json --all Generating CloudFormation template! CloudFormation template written to remote-rule-template.json
New as of version 0.3.11, it is possible to add RuleSet tags to rules that can be used to deploy and test groups of rules together. Rules can belong to multiple RuleSets, and RuleSet membership is stored only in the parameters.json metadata. The deploy, create-rule-template, and test-local commands are RuleSet-aware such that a RuleSet can be passed in as the target instead of --all or a specific named Rule.
A comma-delimited list of RuleSets can be added to a Rule when you create it (using the --rulesets flag), as part of a modify command, or using new ruleset subcommands to add or remove individual rules from a RuleSet.
Running rdk rulesets list will display a list of the RuleSets currently defined across all of the Rules in the working directory
rdk-dev $ rdk rulesets list RuleSets: AnotherRuleSet MyNewSet
Naming a specific RuleSet will list all of the Rules that are part of that RuleSet.
rdk-dev $ rdk rulesets list AnotherRuleSet Rules in AnotherRuleSet : RSTest
Rules can be added to or removed from RuleSets using the add and remove subcommands:
rdk-dev $ rdk rulesets add MyNewSet RSTest RSTest added to RuleSet MyNewSet rdk-dev $ rdk rulesets remove AnotherRuleSet RSTest RSTest removed from RuleSet AnotherRuleSet
RuleSets are a convenient way to maintain a single repository of Config Rules that may need to have subsets of them deployed to different environments. For example your development environment may contain some of the Rules that you run in Production but not all of them; RuleSets gives you a way to identify and selectively deploy the appropriate Rules to each environment.
email me at [email protected] if you are interested in contributing. I'm using the github issues log as my "to-do" list, and I'm also happy to get PR's if you see something you want to fix.
- Michael Borchert - Python version & current maintainer
- Jonathan Rault - Design, testing, feedback
- Greg Kim and Chris Gutierrez - Initial work and CI definitions
- Henry Huang - Original CFN templates and other code
This project is licensed under the Apache 2.0 License
- the boto3 team makes all of this magic possible.