How can I deploy a new cloudfront s3 version without a small period of unavailability?

Question

I am utilizing AWS cloudfront with an S3 origin. I'm using a webpack plugin to cache-bust using chunked hash file names for all of my static files excluding index.html, which I will simply invalidate using the cloudfront feature upon each new release.

I plan on using a jenkins build to run aws s3 sync ./dist s3://BUCKET-NAME/dist --recursive --delete which will swap out the new chunked files as necessary. Then I will overwrite the index.html file to use the new chunked reference. During the few seconds (max) it takes to swap out the old files for new, it is possible that a user will make a request to the website from a region in which cloudfront has not cached the resources, at which point they'll be unavailable because I have just deleted them.

I could not find any information about avoiding this edge case.

Answer 1

Yes, it can happen that a person near a different edge location experience the missing files. To solve this, you need to change the approach of doing new deployments since cache busting and time is unpredictable at request-response level. One commonly used pattern is to keep different directories(paths) for each new deployment in S3 as follows.

For release v1.0
/dist/v1.0/js/*
/dist/v1.0/css/*
/dist/index.html <- index.html for v1.0 release which has reference for js & css in /dist/v1.0 path

For release v1.1
/dist/v1.1/js/*
/dist/v1.1/css/*
/dist/index.html <- index.html for v1.1 release which has reference for js & css in /dist/v1.1 path

After each deployment, a user will receive either the old version(v1.0) or new version(v1.1) of the index.html, which will still working during the transition period until the edge cache is busted.

You can automate the versioning with Jenkins either incrementing the version or using parameterize build plugin.

This will also be useful to do immutable deployments, where in a case of a critical issue, you can rollback to the previous deployments. Apart from that you can configure S3 lifecycle management rules to archive the older versions.