botocove

Botocove

Run a function against a selection of AWS accounts, Organizational Units (OUs) or all AWS accounts in an organization, concurrently with thread safety. Run in one or multiple regions.

Opinionated by default to work with the standard AWS Organization master/member configuration from an organization master account but customisable for any context.

• Fast
• Easy
• Dolphin Themed 🐬

Botocove is a simple decorator for functions to remove time and complexity burden. Uses a ThreadPoolExecutor to run boto3 sessions against AWS accounts concurrently.

Decorating a function in @cove provides a boto3 session to the decorated Python function and runs it in every account requested, gathering all results into a dictionary.

Warning: this tool gives you the potential to make dangerous changes at scale. Test carefully and make idempotent changes! Please read available arguments to understand safe experimentation with this package.

Pre-requisites and Info

An AWS session with sts:assumerole and sts:get-caller-identity access, and accounts that contain a IAM role with trust relationship to the Botocove calling account.

By default, the session is expected to be in an AWS Organization Master or a delegated Organization admin account. You can alter nearly all behaviour of Cove with appropriate arguments

Cove will not execute a function call in the account it's called from.

In the Botocove calling account the minimum IAM requirements are:

• sts:AssumeRole on all of the roles in the target accounts.
• organizations:ListAccounts to run against an AWS organization and capture account metadata.
• organizations:ListChildren to run against specific OUs.

In the organization member accounts:

• A target role that trusts the calling account - for example AWSControlTowerExecution or OrganizationAccountAccessRole role

See the arguments section for how to change these defaults to work with any account configuration, including running without an AWS Organization.

Quickstart

A function written to interact with one boto3 session can now be called with a session from every account and region required by assuming a role in for each account - except the host account you're running from.

For example:

A standard approach: this function takes a boto3 session and gets all IAM users from a single AWS account. You would then manually run it in each account.

import boto3


def get_iam_users(session):
    iam = session.client("iam", region_name="eu-west-1")
    all_users = iam.get_paginator("list_users").paginate().build_full_result()
    return all_users

def main():
    session = boto3.session.Session(profile_name="my_dev_account")
    users = get_iam_users(session)
    print(users) # A single account's IAM users

Now with @cove: a session for every account in the organization is injected by the decorator. A safe example to run as a test!

from pprint import pprint
import boto3
from botocove import cove

@cove()
def get_iam_users(session):
    iam = session.client("iam", region_name="eu-west-1")
    all_users = iam.get_paginator("list_users").paginate().build_full_result()
    return all_users

def main():
    # No session passed as the decorator injects it
    all_results = get_iam_users()
    # Now returns a Dict with keys Results, Exceptions and FailedAssumeRole

    # A list of dictionaries for each account, with account details included.
    # Each account's get_iam_users return is in a "Result" key.
    pprint(all_results["Results"])

    # A list of dictionaries for each account that raised an exception
    pprint(all_results["Exceptions"])

    # A list of dictionaries for each account that could not be assumed into
    pprint(all_results["FailedAssumeRole"])


if __name__ == "__main__":
    main()

Here's an example of a more customised Cove decorator:

@cove(
    target_ids=["123456789101", "234567891011"], # also accepts OU ids!
    rolename="AWSControlTowerExecution",
    raise_exception=True,
    regions=["eu-west-1", "eu-west-2", "us-east-1"],
)
def do_things(session):
    # Cove will return six results of True, 2 accounts * 3 regions
    return True

Arguments

Cove

@cove():

Uses the standard boto3 credential chain to start with, assuming roles in every account required. Defaults to assuming the OrganizationAccountAccessRole in every account in an AWS organization.

Equivalent to:

@cove(
    target_ids=None, ignore_ids=None, rolename=None, role_session_name=None,
    policy=None, policy_arns=None, assuming_session=None, raise_exception=False,
    thread_workers=20, regions=None, partition=None
    )

target_ids: List[str]

A list of AWS account IDs and/or AWS Organization Units IDs to attempt to assume role in to. When unset, attempts to use every available account ID in an AWS organization. When specifying target OU's, all child OUs and accounts belonging to that OU will be collected.

ignore_ids: List[str]

A list of AWS account ID's and OU's to prevent functions being run by Cove. Ignored IDs takes precedence over target_ids. Providing an OU ID will collect all child OUs and accounts to ignore.

The calling account that is running the Cove-wrapped function at runtime is always ignored.

rolename: str

An IAM role name that will be attempted to assume in all target accounts. Defaults to the AWS Organization default, OrganizationAccountAccessRole.

role_session_name: str

An IAM role session name that will be passed to each Cove session's sts.assume_role() call. Defaults to the name of the role being used if unset.

policy: str

A policy document that will be used as a session policy. A non-None value is passed through via the Policy parameter in each Cove session's sts.assume_role() call.

policy_arns: List[PolicyDescriptorTypeTypeDef]

A list of managed policy ARNs that will be used as a session policy. A non-None value is passed through via the PolicyArns parameter in each Cove session's sts.assume_role() call.

assuming_session: Session

A Boto3 boto3.session.Session() object that will be used to call sts:assumerole. If not provided, cove will instantiate one which will use the standard boto3 credential chain.

raise_exception: bool

Defaults to False. Default behaviour is that exceptions are not raised from decorated function. This is due to cove running asynchronously and preferring to resolve all tasks and report their results instead of exiting early.

raise_exception=True will allow a full stack trace to escape on the first exception seen; but will not gracefully or consistently interrupt running tasks. It is vital to run interruptible, idempotent code with this argument as True.

thread_workers: int

Defaults to 20. Cove utilises a ThreadPoolWorker under the hood, which can be tuned with this argument. Number of thread workers directly correlates to memory usage: see here

regions: List[str]

If not provided, Cove will respect your profile's default region via the boto credential chain. If provided, Cove will run the decorated function in every region named.

You can get all regions with:

regions = [
    r['RegionName'] for r in boto3.client('ec2').describe_regions()['Regions']
    ]

partition: str

If not provided, Cove will use the AWS partition of your profile in constructing the ARN for the role to assume in all target accounts.

external_id: str

Defaults to None. An external id that will be passed to each Cove session's sts.assume_role() call.

CoveSession

Cove supplies an enriched Boto3 session to each function called. Account details are available with the session_information attribute.

Otherwise, it functions exactly as calling boto3 would.

@cove()
def do_nothing(session: CoveSession):
    print(session.session_information) # Outputs a dict of known information
    # This function runs no boto3-specific API, but does test that a role
    # can be assumed

Return values

Wrapped functions return a dictionary. Each value contains List[Dict[str, Any]]:

{
    "Results": results:
    "Exceptions": exceptions,
    "FailedAssumeRole": invalid_sessions,
}

An example of cove_output["Results"]:

[ # A list of dictionaries per account called
    {
    'Id': '123456789010',
    'Email': '[e-mail]',
    'Name': 'account-name',
    'Status': 'ACTIVE',
    'AssumeRoleSuccess': True,
    'Result': wrapped_function_return_value # Result of wrapped func
    }
]

Is botocove thread safe?

botocove is thread safe, but number of threaded executions will be bound by memory, network IO and AWS api rate limiting. Defaulting to 20 thread workers is a reasonable starting point, but can be further optimised for runtime with experimentation.

botocove has no constraint or understanding of the function it's wrapping: it is recommended to avoid shared state for botocove wrapped functions, and to write simple functions that are written to be idempotent and independent.

Boto3 Session objects are not natively thread safe

and should not be shared across threads. However, botocove is instantiating a new Session object per thread/account and running decorated functions inside their own closure. A shared client is used from the host account that botocove is run from (eg, an organization master account) - clients are threadsafe and allow this.

boto3 sessions have a significant memory footprint: Version 1.5.0 of botocove was re-written to ensure that boto3 sessions are released after completion which resolved memory starvation issues. This was discussed here: https://github.com/connelldave/botocove/issues/20 and a relevant boto3 issue is here: https://github.com/boto/boto3/issues/1670

botocove?

It turns out that the Amazon's Boto dolphins are solitary or small-group animals, unlike the large pods of dolphins in the oceans. This killed my "large group of boto" idea, so the next best idea was where might they all shelter together... a cove!