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 ๐ฌ
Botocove is a simple decorator for functions to remove time and complexity
burden. UsesThreadPoolExecutor
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.
An IAM user with sts:assumerole
privilege, and accounts that have a trust
relationship to the IAM user's account.
By default, the IAM user is expected to be in an AWS Organization Master 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.
Default requirements are:
In the organization master account:
- IAM permissions
sts:assumerole
,sts:get-caller-identity
andorganizations:list-accounts
In the organization member accounts:
See the arguments section for how to change these defaults to work with any account configuration, including running without an AWS Organization.
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
@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,
org_master=True, thread_workers=20, regions=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
.
org_master
: bool
Defaults to True. When True, will leverage the Boto3 Organizations API to list
all accounts in the organization, and enrich each CoveSession
with information
available (Id
, Arn
, Name
, Status
, Email
). Disabling this and providing
your own full list of accounts may be a desirable optimisation if speed is an
issue.
org_master=False
means only Id
will be available to CoveSession
.
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']
]
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
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': '[email protected]',
'Name': 'account-name',
'Status': 'ACTIVE',
'AssumeRoleSuccess': True,
'Result': wrapped_function_return_value # Result of wrapped func
}
]
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: connelldave#20 and a relevant boto3 issue is here: boto/boto3#1670
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!