Hacker News Reader

Comment by the_mitsuhiko | Hacker News Reader

> I think there's a reasonable middle ground-point between having feature flags in a JSON file that you have to redeploy to change and using an (often expensive) feature flags as a service platform: roll your own simple system.

The middle ground is a JSON file that is copied up and periodically refreshed. We (Sentry) moved from a managed software to just a YAML file with feature flags that is pushed to all containers.

The benefit of just changing a file is that you have a lot of freedom of how you deal with it (eg: leave comments) and you have the history of who flipped it and for which reason.

View on HN

How do you push the files to all of your containers? I’ve done this in the past with app specific endpoints but never found a solution I liked with containers.

We keep a JSON blob in Google Secret Manager for our flags. The service running in the container will reload the secret anytime it changes

Ah that’s a super nice feature. I’m mostly familiar with AWS who don’t have a neat way of doing this, you end up with a bespoke solution either with lambdas pushing to shared volumes or just polling s3 for updates.

If you are using kubernetes, you can mount the secret/ConfigMap as a volume and it will be updated automatically when changes occur. Then your application merely watches the file for updates.

Being on AWS, using EKS feels like overkill when you're talking $75/month just for having it managed by AWS. This doesn't work with ECS, unfortunately, or if you're just running docker on EC2.

AWS has a native service for this called AppConfig and has agents that can pull and cache flag values so your services only need to make localhost requests.

AH nice, I was not aware of this. Thanks (It is expensive, though...)

Expensive?? Really? One of the cheapest services around.

$0.0000002 per configuration request

Depending on when you’re evaluating it’s a per request overhead. You might/provably have multiple flags per request. Compared to a lambda invocation that pushes a config file to every container if it changes, it’s expensive.

We currently persist the feature flag config in a database where the containers pull it from. Not the optimal solution but that was a natural evolution from a system we already had in place.

Reply on news.ycombinator.com