Triggering a Cloudflare Cache Purge from S3 Object Events

The Challenge

I’ve have been looking at some cost and performance improvements in some of my infrastructure. This particular environment is essentially a handful of websites which use some shared static assets that reside in an S3 bucket think which is used by, and

These websites are currently all running through Cloudflare’ CDN service, but this assets cache is fronted by AWS Cloudfront.

user flow diagram before

With our Cloudflare CDN services we have implemented a load of optimisations; including the Cloudflare Argo service (which is ace, but a subject for another blog).

Having now completed the migration from AWS Cloudfront to Cloudflare to the user flow looks more like this:

user flow diagram after

With the assets in the S3 bucket we found a situation where by the marketing teams upload files and sometimes; these uploads contain fixes or edits which need to be changed on the website NOW. Waiting for the normal TTL to expire just wont work (typos, bad information etc); historically they would just log in to AWS, locate the distribution and run a cache invalidation. Access was govered by IAM and limited to this specific action on this specific distribution.

With Cloudflare’s access model being a bit shit limited it’s all or nothing. You can create users based on the account which will have access to all zones within it.

cloudflare user perms 1

Your choices are to have a separate account for each domain/zone you manage if you want less access for users but a worse management experience, or you have the related zones in the same account but have an overly permissive accounts model

Whilst there is a ‘purge_cache’ privilage option it grants this access to all domains/ones in the account; which is not least-privilege and not something I want to deal with (Cloudflare are making improvements in this area apparently).

cloudflare user perms 1

The Solution

I needed a solution which would purge the cache when specific objects or types of objects are uploaded to S3, ideally without any user interaction whatsoever.

I created a fairly simple Lambda function which is triggered by S3 events, it takes in the Cloudflare API details from an AWS SSM Parameter Store value. Given the S3 event we can assemble the URL, generate a POST request and send a request to the Cloudflare API to purge this specific item (code at the bottom of this post).

In the past this S3 event would have had to go on an AWS SQS queue which was then polled by the Lambda function. Since March this year we’ve been able to trigger a Lambda function directly from an S3 object event eliminating some complexity in this particular situation.

Given my requirements it still might be better to use an SQS queue in the middle to ensure event delivery and provide some ordering and possibly a little more error handling for failed processing; but as this is just a case of picking up that a S3 object key has changed then firing its path over to an external API it seemed unnessessary. This was also a pretty fun way to try out this new-ish functionality.

It might turn out that when loads of files are uploaded we are loosing some and actually an SQS queue makes more sense to batch things up. I might also switch to using tags on the objects in the Cloudflare space to make purging a bit more efficient.

Script to invalidate cloudflare cache objects based on s3 object triggers. Assumes the bucket name is the same as the target URL

Environmental Variables:

    Type: "String"
    Usage: [optional] Cloudflare API KEY.

    Type: "String"
    Usage: [optional] SSM Path for the Cloudflare API KEY.

    Type: "String"
    Usage: SSM Path for the Cloudflare API KEY.

    Type: String
    Usage: Cloudflare Zone ID for the target of the invalidations

    Type: String, comma seperated list
    Usage: A comma seperated list of extensions to invalidate cache for, default is everything
    Default: *

    Type: String, comma seperated list
    Usage: A comma seperated list of extensions NOT to invalidate cache for, default is nothing
    Default: empty

  Designed to be run as a lambda and triggered by an S3 event.
  If running manually, you'll need to set the environmental variables first

import boto3
import json
import sys
import urllib3

from os import environ

def invalidate_cf_caches(invalidations):
  ssm = boto3.client('ssm')

    api_key = ssm.get_parameter(Name=environ.get("CLOUDFLARE_API_KEY_SSMPATH"), WithDecryption=True)

  elif "CLOUDFLARE_API_KEY" in environ:
    api_key = environ.get("CLOUDFLARE_API_KEY")


  api_email = environ.get("CLOUDFLARE_API_EMAIL") if "CLOUDFLARE_API_EMAIL" in environ else sys.exit("CLOUDFLARE_API_EMAIL is missing")
  zone_id = environ.get("CLOUDFLARE_ZONE_ID") if "CLOUDFLARE_ZONE_ID" in environ else sys.exit("CLOUDFLARE_ZONE_ID is missing")

  url = "" + zone_id +"/purge_cache"

  payload = json.dumps({
    "files": invalidations

  headers = {
    'X-Auth-Email': api_email,
    'X-Auth-Key': api_key,
    'Content-Type': 'application/json',

  http = urllib3.PoolManager()

  response = http.request('POST',
    body = payload,
    headers = headers,
    retries = False)

  # is a bytes array, decode it to a json object making it somewhat useful
  response_usable = json.loads('utf8').replace("'", '"'))
  response_usable["invalidations"] = invalidations

  return response_usable

# Lambda Handler
def lambda_handler(event, context):
  invalidations = []
  included_extensions = [] if not "INCLUDED_EXTS" in environ else [x.strip() for x in environ.get("INCLUDED_EXTS").split(",")]
  excluded_extensions = [] if not "EXCLUDED_EXTS" in environ else [x.strip() for x in environ.get("EXCLUDED_EXTS").split(",")]

  for record in event["Records"]:
    # If the current file has an extension not permitted by the included extensions filter, skip to the next record
    if len(included_extensions) > 0 and record['s3']['object']['key'].split(".")[-1] not in included_extensions:
      print(record['s3']['object']['key'], "does't have an extension in the inclusion filter and will be skippedun")
      # The INCLUDE_EXTS environmental variable doesnt have a matching file type

    if len(excluded_extensions) > 0 and record['s3']['object']['key'].split(".")[-1] in excluded_extensions:
      print(record['s3']['object']['key'], "has a excluded extension, and will be skipped")
      # The EXCLUDE_EXTS environmental variable has a matching file type to be excluded

    invalidations.append("https://" + record['s3']['bucket']['name'] + "/" + record['s3']['object']['key'])

  if len(invalidations) > 0:
    invalidation_response = invalidate_cf_caches(invalidations)
    invalidation_response="All updated S3 object keys were matched on the exclusion list or not explicitily included"

  return invalidation_response

Unified authentication with AWS Cognito!

Attempts to minimise the number of credentials required is becoming more and more common, particularly with the adoption of social logins. Many well known services are backing off their authentication mechanisms to Google, Facebook, GitHub etc and passing a session token back to the initial site.

This greatly reduces the number of passwords someone needs to remember and allows things like MFA to be implemented easily without small sites/service having to invest the engineering time in building out their authentication mechanisms and maintaining them.

The downside of these authentication providers is they become a target, if an attacker gains access to say a Facebook account, all the 3rd party services that where linked to that account are now compromised too. The obvious solution is MFA, strong passwords and that the provider is prioritising security; it’s still better than the alternative where sites are compromised and it turns out passwords were stored in clear text along with credit card details.

social login

Recently I came across an application which needed to authenticate a set of users from 3 separate companies, (owner/client’s IT and two external companies). The requirements for the access boils down to everyone must have a named account for auditing but we don’t want to have to create everyone a new user account when the start on the project. The reasoning for this is as people start and leave it means three companies need to update their new starter and leavers processes and it’s hard enough getting one to do it.

All three companies have their own Azure AD environments set up already and this would be the ideal solution if we can get the application to redirect to Azure. The first and most obvious problem with this is the application would need to support 3 different Azure AD connections which isn’t something it can handle.

Enter AWS Cognito!

AWS cognito can be configured to act like an authentication concentrator, in this case each company creates an Azure AD Enterprise Application for the Cognito pool and we create an configuration for the target application. Cognito can be configured to use pretty much anything that supports OpenID Connect or SAML (or Facebook, Google or Amazon’s own store users). Additional users can be created within Cognito itself for adhoc or service/api users if needed.


Each of the companies will create an Azure Enterprise Application for the Cognito deployment as a SAML target and allow their relevant users access to said Enterprise App. This isn’t exactly a complicated process to complete. Once the created, Azure will generate an XML document which needs to be added as an Identity Provider in Cognito and thats it. As long as a user is a member of a group with access to authenticate against that Enterprise Application Cognito will respect it.

Our target application is then setup to use Cognito over OpenID Connect. the application login redirects the user to Cognito which can then direct users to their Azure tenant. The downside to this approach is that the owner/client of the application must trust the other companies to have sufficient security and policies around access.