Nested Stacks are a great way to deploy your infrastructure in a modular fashion. Isolating resources into logical groups allows us to keep our CloudFormation scripts small, limit blast radius of changes, and provide an easy way to manage various resources in more specific templates. There are a few inherit challenges that we face when using this great technology, and the one we are focusing on today is version control of our child templates.
As the URL requirement for nested stacks requires a(n) S3 URL, making changes can quickly become a hassle when there are several child templates that need to be maintained. Enter a great quick-start by AWS that gets us 80% of the way there. This quick-start will create Lambda functions that will take the contents of your git repo and upload a ZIP of them to S3. This is a great solution, but a ZIP file might not fit your organization's needs, especially if we want to use the files directly via URL.
We at Onica have faced this challenge numerous times, and as a result have modified the AWS provided scripts to actually deposit the contents of your repo while maintaining the original directory structure found with your repo. Our method involves taking the code from the AWS provided scripts and hosting them within your own S3 bucket, allowing us to modify the Lambda functions at our leisure. Use the following steps to create a stack that makes the necessary modifications.
Note: These steps can be followed to illustrate the process. You can also simply create the S3 bucket and use the Cloudformation template at the end of this blog.
Using the quick-start guide, follow the steps for configuration, ensuring you configure for the Git pull endpoint strategy, not the Zip download endpoint strategy. This initial configuration will consist of setting up the webhook in your repo, configuring an SSH keypair, and deploying the Lambda functions. Once this is in-place, be sure to do a test to verify everything is working correctly.
While we won’t dive into detail on these steps as they are outlined in the quick start guide, the basics are creating a bucket for your repo objects to live, providing the IP ranges for your preferred repo, and designating the buckets to pull the quick-start files from. This will take 15-20 minutes, and all steps are well-defined within the guide.
IMPORTANT: Do not modify the AWS Quick Start Configuration parameters at this time. We will do so later.
Once you have run the template for the first time, all of the necessary script will be copied to a bucket within your account. We want to create our own bucket with a friendlier name so we can house and modify the code. Our new bucket should have a simple name, something like “[companyname]-quickstarts”.
aws s3api create-bucket --bucket [BucketName --region [Region] --create-bucket-configuration LocationConstraint=[Region]
The initial setup will have created a bucket in your account with the following pattern: [stackname]-lambdazipsbucket-[randomhash]. Take the contents of this bucket(including folder structure) and copy them to the custom bucket that was just created.
The following command can be used to quickly move the files:
aws s3 cp s3://[SourceBucketName]/ s3://[TargetBucket]/ --recursive
You should now have all of the lambda scripts in your bucket.
There is only one Lambda function that needs to be modified. It can be found at s3://[BucketName]/git2s3/latest/lambdas/GitPullS3.zip. Pull down a copy of this zip file, unzip it, and replace the file lambda_function.py with the following script:
from pygit2 import Keypair,credentials,discover_repository,Repository,clone_repository,RemoteCallbacks
from boto3 import client,resource
import os,stat
import shutil
from zipfile import ZipFile
from ipaddress import ip_network, ip_address
import json
import logging
import hmac
import hashlib
### If true the function will not include .git folder in the zip
exclude_git=True
### If true the function will delete all files at the end of each invocation, useful if you run into storage space constraints, but will slow down invocations as each invoke will need to checkout the entire repo
cleanup=False
key='enc_key'
logger = logging.getLogger()
logger.setLevel(logging.INFO)
logger.handlers[0].setFormatter(logging.Formatter('[%(asctime)s][%(levelname)s] %(message)s'))
logging.getLogger('boto3').setLevel(logging.ERROR)
logging.getLogger('botocore').setLevel(logging.ERROR)
s3 = client('s3')
kms = client('kms')
s3r = resource('s3')
def write_key(filename,contents):
logger.info('Writing keys to /tmp/...')
mode = stat.S_IRUSR | stat.S_IWUSR
umask_original = os.umask(0)
try:
handle = os.fdopen(os.open(filename, os.O_WRONLY | os.O_CREAT, mode), 'w')
finally:
os.umask(umask_original)
handle.write(contents+'\n')
handle.close()
def get_keys(keybucket,pubkey,update=False):
if not os.path.isfile('/tmp/id_rsa') or not os.path.isfile('/tmp/id_rsa.pub') or update:
logger.info('Keys not found on Lambda container, fetching from S3...')
enckey = s3.get_object(Bucket=keybucket,Key=key)['Body'].read()
privkey = kms.decrypt(CiphertextBlob=enckey)['Plaintext']
write_key('/tmp/id_rsa',privkey)
write_key('/tmp/id_rsa.pub',pubkey)
return Keypair('git','/tmp/id_rsa.pub','/tmp/id_rsa','')
def init_remote(repo, name, url):
remote = repo.remotes.create(name, url, '+refs/*:refs/*')
return remote
def create_repo(repo_path, remote_url, creds):
if os.path.exists(repo_path):
logger.info('Cleaning up repo path...')
shutil.rmtree(repo_path)
repo = clone_repository(remote_url, repo_path, callbacks=creds )
return repo
def pull_repo(repo, remote_url, creds):
remote_exists = False
for r in repo.remotes:
if r.url == remote_url:
remote_exists = True
remote = r
if not remote_exists:
remote = repo.create_remote('origin',remote_url)
logger.info('Fetching and merging changes...')
remote.fetch(callbacks=creds)
remote_master_id = repo.lookup_reference('refs/remotes/origin/master').target
repo.checkout_tree(repo.get(remote_master_id))
master_ref = repo.lookup_reference('refs/heads/master')
master_ref.set_target(remote_master_id)
repo.head.set_target(remote_master_id)
return repo
def zip_repo(repo_path,repo_name):
logger.info('Creating zipfile...')
zf = ZipFile('/tmp/'+repo_name.replace('/','_')+'.zip','w')
for dirname, subdirs, files in os.walk(repo_path):
if exclude_git:
try:
subdirs.remove('.git')
except ValueError:
pass
zdirname = dirname[len(repo_path)+1:]
zf.write(dirname,zdirname)
for filename in files:
zf.write(os.path.join(dirname, filename),os.path.join(zdirname, filename))
zf.close()
return '/tmp/'+repo_name.replace('/','_')+'.zip'
def push_s3(filename,repo_name,outputbucket):
s3key='%s/%s' % (repo_name,filename.replace('/tmp/',''))
logger.info('pushing zip to s3://%s/%s' % (outputbucket,s3key))
data=open(filename,'rb')
s3.put_object(Bucket=outputbucket,Body=data,Key=s3key)
logger.info('Completed S3 upload...')
def sync_to_s3(path, repo_name, bucketname):
for root,dirs,files in os.walk(path):
for filename in files:
path_and_file = os.path.join(root,filename).replace('/tmp/','')
logger.info("processing %s", path_and_file)
s3.upload_file(os.path.join(root,filename),bucketname, path_and_file)
logger.info('File uploaded to https://s3.%s.amazonaws.com/%s/%s' % (
os.environ['AWS_REGION'], bucketname, path_and_file))
def empty_bucket(bucketname):
bucket = s3r.Bucket(bucketname)
bucket.objects.all().delete()
def lambda_handler(event,context):
print(json.dumps(event))
keybucket=event['context']['key-bucket']
outputbucket=event['context']['output-bucket']
pubkey=event['context']['public-key']
### Source IP ranges to allow requests from, if the IP is in one of these the request will not be chacked for an api key
ipranges=[]
for i in event['context']['allowed-ips'].split(','):
ipranges.append(ip_network(u'%s' % i))
### APIKeys, it is recommended to use a different API key for each repo that uses this function
apikeys=event['context']['api-secrets'].split(',')
ip = ip_address(event['context']['source-ip'])
secure=False
for net in ipranges:
if ip in net:
secure=True
if 'X-Gitlab-Token' in event['params']['header'].keys():
if event['params']['header']['X-Gitlab-Token'] in apikeys:
secure=True
if 'X-Git-Token' in event['params']['header'].keys():
if event['params']['header']['X-Git-Token'] in apikeys:
secure=True
if 'X-Gitlab-Token' in event['params']['header'].keys():
if event['params']['header']['X-Gitlab-Token'] in apikeys:
secure=True
if 'X-Hub-Signature' in event['params']['header'].keys():
for k in apikeys:
if hmac.new(str(k),str(event['context']['raw-body']),hashlib.sha1).hexdigest() == str(event['params']['header']['X-Hub-Signature'].replace('sha1=','')):
secure=True
if not secure:
logger.error('Source IP %s is not allowed' % event['context']['source-ip'])
raise Exception('Source IP %s is not allowed' % event['context']['source-ip'])
try:
repo_name = event['body-json']['project']['path_with_namespace']
except:
repo_name = event['body-json']['repository']['full_name']
try:
remote_url = event['body-json']['project']['git_ssh_url']
except:
try:
remote_url = 'git@'+event['body-json']['repository']['links']['html']['href'].replace('https://','').replace('/',':',1)+'.git'
except:
remote_url = event['body-json']['repository']['ssh_url']
repo_path = '/tmp/%s' % repo_name
creds = RemoteCallbacks( credentials=get_keys(keybucket,pubkey), )
try:
repository_path = discover_repository(repo_path)
repo = Repository(repository_path)
logger.info('found existing repo, using that...')
except:
logger.info('creating new repo for %s in %s' % (remote_url, repo_path))
repo = create_repo(repo_path, remote_url, creds)
pull_repo(repo,remote_url,creds)
empty_bucket(outputbucket)
sync_to_s3(repo_path,repo_name=repo_name, bucketname=outputbucket)
if cleanup:
logger.info('Cleanup Lambda container...')
shutil.rmtree(repo_path)
shutil.rm(zipfile)
shutil.rm('/tmp/id_rsa')
shutil.rm('/tmp/id_rsa.pub')
return 'Successfully updated %s' % repo_name
Once replaced, repack the zip file with the same name and upload to your S3 bucket, replacing the existing GitPullS3.zip file.
Now we need to redeploy our Cloudformation Template to source the code from our S3 bucket as opposed to the AWS Quick Start bucket. The reason for the redeploy as opposed to Stack Update is that the existing IAM roles will cause a CopyObject error as we are switching to a new S3 bucket as the source for the configuration files.
-
Go to Cloudformation and delete the existing stack.
-
Create a new stack with our now modified template.
-
Update the fields to now reflect our new bucket for sourcing the code.
[INSERT IMAGE HERE]
-
Now continue through to creation of your stack with the new values.
-
Configure your repo of choice with the webhook and access key generated by the new stack.
-
Finally, make a commit to your repo to test. You should see your repo cloned into S3.
While the steps above will get you there, it would be even easier to use a pre-created template with all modifications in place. Please find the Cloudformation JSON file and python script on our public github.#
Happy coding!