Devstack: Razorpay's cloud native development ecosystem
- Client only, developer friendly stack for running cloud workloads
- Ability to define a collection of dependent micro services(service fleet) in a composable fashion
- Ability to deploy service fleet and associated cloud infrastructure using a single command
- Build, Test, debug inside kubernetes with hot reloading capability
- Create unified blueprints for your deployment stack using an ephmeral infrastructure
- Service Level/Feature Level routing via header propogation
At razorpay, we run all our workloads on kubernetes. Like any other mature organization, we have an involved CI/CD practice with extremely sophisticated pipelines. While this works great for all production and pre-production workloads, we have been noticing over a period of time a bunch of development challenges. In essence, the goal is to Simplify developer workflow and reduce the time taken to rollout features independently. Devstack, offers a set of tools to help build and develop code on the individual developer's laptop, as if they are working on a cloud environment.
In a nutshell: "Its a client based development tool for building cloud native applications on kubernetes"
Provisioning Service Fleet
Hot sync local code to cloud
As we mentioned earlier, our solution is slightly opinionated and requires the following stack:
- Cloud Provider : AWS(Note: Our custom infra helm hooks are all designed for AWS. However, it can be extended. See details below)
- Kubernetes: Requires verion 1.15+
- Traefik: 2.0+ to be deployed on the kubernetes cluster above. Please see the official installation instructions
- Helm: 3.0+
- LocalStack: To be deployed on kubernetes cluster above. Refer to LocalStack
- Kube Janitor: Deploy Kube Janitor
Other requirements (For hot-reload) If you are using a loosely typed language like php / python etc, then you can safely skip this section. For static languages like golang, java, nodejs etc, please refer below:
- Golang: CompileDaemon
- NodeJs: Nodemon
- Java: GradleDaemon or MavenDaemon
Please look into the docs folder for specific examples. Some examples are listed below:
- Example of a base application and a feature release - Helmfile Based Deployment
- Example of application with hot-reload to running container - Golang app with hot-reload / CompileDaemon
- Example of a SQS Provisioning with helm hooks - Example SQS Producer Consumer App
- Streamlined Dev Workflow: Provide a streamlined workflow and faster merges to
main
ormaster
branches. - Consistent Environment: Provide a seamless consistent environment across dev, stage, pre-prod and production environments
- Faster Feedback: Reduce time to write and build containerized applications. Enable faster feedback loop on local development environment
- Remove vendor lock-in (rely on OSS practically)
- Kubernetes native (At the moment, we don't have extensions for non K8s solutions). Our environment is kubernetes native
- Hassle Free onboarding - minimal changes to application and development lifecycle
- Cost Effective - We should eventually be able to bill developers and teams on usage patterns.
- Slightly Opinionated - This isn't a PaaS offering at the momemt. And hence can be extended and deployed on any native kubernetes installation
- Build, Test and Deploy applications from laptop directly into kubernetes using simple CLI tools
- Ship code to remote container without tunneling: Sync files into container directly using File Sync(using Devspace)
- Ability to provide hot reloading of apps : sync directly into the containers without restart(e.g. CompileDaemon for statically typed languages)
- Support out-of-the-box support using existing available helm hooks
- Support for extensible custom Helm Hooks that handles provisioning of AWS infrastructure components using LocalStack
- Declarative-ly define service and service dependencies using Helmfile
- Traffic routing to right upstream. Uses traefik 2.0 IngressRoute
- Ability to selectively route traffic to different upstream. Done via opentelemetry context/header propagation
- Ability to expose preview URL for all services
- Automatic timed cleanup(
ttl
) of resources using Kube Janitor
Please refer to the Architecture Overview for entire details on the devstack architecture.
We have provided a collection of custom helm hooks for AWS and kubernetes specific workloads. All of these are extensible. Please refer to the documentation of the helm hooks
Please file an issue on this repo using the following labels: Clarification, Feature, Bug
Please refer to the Contribution Guide