[RFC] Docker Compose for OpenReplay installation

[rjshrjndrn]

Hi,

Some users from the community asked whether we can use docker-compose for the OpenReplay installation. As of now, OpenReplay is Kubernetes native application, and for simple installations, we've got an installation script that installs k3s(a single node Kubernetes instance) in the local machine and provision OpenReplay in that, and you can start using OpenReplay immediately.
So if you don't want to fine-grain control over the orchestrator from day one (in this case Kubernetes), you can update your customizations in vars.yaml and do cd openreplay/scripts/helmcharts/ && ./openreplay-cli -I for customizing your installation, like increasing memory/cpu limits or updating an image with custom changes, etc.

These are the following assumptions we have for OpenReplay.

1. Openreplay can't be run on the local machine, unless it's for development purposes, to test the features, as we need FQDN and SSL. So a remote server ( minimum 2 core 8G) is always advisable.
2. For production use, we'll never use docker-compose, so it's better to familiarize yourself with the setup from day one. As the installation is not complicated, just running a script, you can figure things out as you advance.
3. Utilize Kubernetes native plugins/apps to make day to operations easy, for example, SSL certificate rotation.
4. Use helm to template the variables, and keep all your customizations in a single file, so that you can reproduce OpenReplay installation in any cluster, resulting in expected behavior.

We want to know what use cases docker-compose will help you in addition.

Please comment down your usecases below

[andreialecu]

For AWS, CDK stacks are all the rage these days (https://github.com/aws/aws-cdk). You can use CDK to define the infrastructure in TypeScript, Python, or other languages.

For our CDK stacks, we usually work with ECS clusters. ECS is based on docker images and is usually more cost-effective than EKS.

It's possible to use a fleet of “spot instances” with ECS, which are up to 70% cheaper than regular instances. ECS handles recycling and starting new ones if necessary, so there’s never any downtime.

This would be easy to set up if a stateless OpenReplay docker image was made available that could plug into an RDS PostgreSQL database and a separate S3 deployment.

One other nice thing to have would be to be able to use environment variables instead of modifying the configuration file. Not sure if this is already possible or not.

With environment variables, the docker image can stay static, and the user could just provide the database connection string and S3 bucket details as env vars. Upgrades would also be seamless by just swapping a new docker image in place.

[sam17896]

For GCP, Cloud run can used for serverless deployment, application can be deployed using docker image and recordings can be stored on google cloud storage, also the cloud sql can be used as external db option, cloud run manage instance replication and resource allocation by itself, also the dependency of managing a VM or EC2 instance (server patching etc) is not required on cloud run.

[ahadsheriff]

Another cool thing about deploying OpenReplay in Google Cloud Run and AWS App Runner is that the services would handle SSL certificate management and reduce overhead complexity of scaling the service. It's simply an easier deployment solution, especially for those of us who don't want to spend time managing an entire Kubernetes cluster.

[astrot1988]

docker swarm

[smsteel]

we want to use it as a self-hosted service at our own infrastructure and we don't have k8s yet

[rjshrjndrn]

For selfhosting, you only need a VM. everything else is abstracted.

you can get started with

1. download the cli sudo wget https://raw.githubusercontent.com/openreplay/openreplay/main/scripts/helmcharts/openreplay-cli -O /bin/openreplay ; sudo chmod +x /bin/openreplay
2. Install OR: openreplay -i <domain_name>

Ref: https://docs.openreplay.com/en/deployment/deploy-ubuntu/