Flynn Spec

General Features

• Manages services in containers across a cluster
• HTTP RESTish API allows easy management of services/containers
• Comes with buildpacks and “git push” deployment
• Open source (BSD 3-clause) and runs on any infrastructure (AWS/Public Cloud, Private Cloud, Bare Metal)
• Designed with twelve-factor apps in mind
• Services are UNIX processes, totally language-agnostic
• Complex deployment options like staged rollouts, continuous deployment, etc. can be built by using the API
• Backing services (with persistent disk) may be deployed and managed using the same system
  • Provides HA and provisioning hooks
• HTTP and TCP routing/load-balancing provided
• The system itself is fully containerized, with few (if any) components not running in containers
• Systems utilizing containers (for example CI) can be built using the API

MVP Features

• API
  • CRUD /services
    • container management
      • scale up/down
      • push new image
      • stream logs
• Scheduler
• git push buildpack image builder
• HTTP router
• Command-line client

Final Product Features

• TCP routing (with PROXY protocol)
• Disk-backed service support
  • Pin container to node for RAID/EBS usage
  • HA master/slave setup/failover hooks
  • DB provisioning hooks (create/destroy database)
• Run/attach (“heroku run”)
• User accounts with basic ACLs
• HA for routing, graceful failure for scheduler, etc.

Reach Goals

• AMI for easy deployment
• Prebuilt HA setups for common databases (Redis, Postgres, Riak, Mongo, MySQL)
• Monitoring system integration
• Graphite/metrics integration
• Service registry
• Professionally designed web dashboard
• More web dashboard stuff (databases, repo viewer, etc)
• Continuous integration tool
• Log Aggregation (compatible with open source analysis tools)
• ACLs/quotas (user, team, etc)
• Autoscaling
• HA control plane
• Scheduler auto-failover
• Priority (run Hadoop when there is cluster availability)
• Hadoop/Spark setups
• Ceph

Benefits

• Painless scaling, just add more nodes
• Painless deployment, just git push (“internal Heroku”)
• Ops provides PaaS as a product to software engineering, no messing around with configuration management scripts to deploy tiny apps
• Self-serve access for developers makes deploying internal services and one-off projects completely painless for everyone
• Easily deploy existing open source applications
• Saves money by not using one or more VM per service component
• Best practices included for HA, log aggregation
• Provides a standardized environment for the Ops team to manage
• Simple and composable vs existing Open Source PaaS products