Automating Cloudfront invalidations in an continous deployment setup
This previous post worked through the process of automating deployment of a Hugo blog to an Amazon AWS environment. Part of that included setting up a Cloudfront distribution to act as a content distribution network in front of the S3 bucket.
Using Cloudfront as a CDN means that the actual content is replicated out into the CDN’s edge nodes, and will be served from there rather than hitting the “real” backend. The tradeoff is that changes to the “real” content will take time to replicate out to the edge nodes, so it can be a while before they show up. Not ideal if you want to get new or updated content out there quickly.
A workaround for this delay is to create an “invalidation” - basically telling Cloudfront that the cached versions of specific files are now stale, and need to be updated.
There’s a cost for creating invalidations, so they’re not something that you want to do multiple times an hour - but for getting the latest version of a site out there, it’s not too onerous.
You can create invalidations manually - but rather than forgetting and then sit around wondering why nothing has changed, a more robust process is to add them to the deployment stage of your CI workflow.
The prerequisites are:
- you’ve created a Cloudfront distribution, and you’ve got a note of its ID
- the AWS user that runs the deployment stage has the
CloudFrontFullAccesspermission policy attached
- you’ve got a CI/CD workflow set up on Gitlab as per the original post. The basic principles will also apply to other CI services such as Github actions.
The overall process
The overall process is as follows:
- you push changes to Gitlab, which triggers the build/deploy pipeline
- the build job handles building the files to be pushed to AWS
- the deploy job copies those built files up to S3
- at the end of the deploy job, it creates a Cloudfront invalidation with the AWS CLI tool that’s running in the deployment container. That invalidation will force the new and updated files to be replicated immediately.
Configuring the Gitlab settings
To keep the deployment configuration and the parameters (such as access keys, secrets and Cloudfront distribution ID) separate, you’ll use a Gitlab environment variable which will be substituted into the config at runtime.
In the Gitlab project, select the
CI/CD option from the
Settings area in the left-hand sidebar, and add a new
Key is up to you - I use
value is the Cloudfront distribution ID from the prerequisites.
Updating the jobs
Once you’ve got the Cloudfront distribution ID saved as a variable, you can use it in the
.gitlab-ci.yml configuration file.
prod_deploy stage we’re going to add a third script step, which will create a new Cloudfront invalidation as soon as the built files have been copied to S3.
Add the following below the S3 script:
aws cloudfront create-invalidation --distribution-id $PROD_CLOUDFRONT_DIST --paths "/*"
This creates an invalidation for the distribution ID that’s stored in the
$PROD_CLOUDFRONT_DIST variable, and uses a wildcard path to invalidate all files in the distribution.
(Technically, you could scope the wildcard down a bit to save on potential costs (you get 1000 file invalidations free per month, and they cost $0.005 thereafter) - but using the
* wildcard means that you won’t accidentally miss less-obvious changes.)
The complete stage should now look like:
prod_deploy: stage: prod_deploy image: python:latest script: - pip install awscli - aws s3 cp ./public_html s3://$PROD_S3_BUCKET_NAME/ --recursive --acl public-read - aws cloudfront create-invalidation --distribution-id $PROD_CLOUDFRONT_DIST --paths "/*" only: - master
Checking the results
If you now trigger a production deployment by pushing to the
master branch, you’ll see an extra step being logged once all the file copies have taken place:
If you reload your live site after a minute or so, you should see your changes immediately viewable.