Open source and 6 commercial distrosGlobal 2000 focus
Launched 2011
Cloud Native Application Platform
A single API for managing applications on 4 infrastructures
32,000 meetup members2,100 committers
50+ foundation companies
Major enterprise adoption:Huawei running 5,000+ apps
GE next-gen Internet of Things platformBaidu has 700+ CF devs
$ cf scale
1. Designed for openness and extensibility
2. Flexible cloud primitives and processes
3. A platform that can keep promises
What Makes This Runtime Interesting?
Develop and administer thousands of cloud native applications with heterogeneous workloads and
architectures
The Problem
1. Design
2. Build3.
Configure
4. Deploy
5. Run
1. Provision
2. Authorize
3. Integrate
4. Monitor5. Audit
6. Scale
7. Recover
8. Upgrade
Dev Loop
Ops Loop
• Role-based to resource access • Run code on demand• Coordinate cross-service configuration• Route public requests• Read and write persistent data• Record internal and external events
• Isolate resources and failures
• Measure performance/health
• Detect and determine failure
• Failure recovery• Work tomorrow• Add and remove resources
Runtime Capabilities
Core Runtime Concepts
CellsContainer Execution
BrainContainer Scheduler & Health Monitor
BBSShared Runtime State
Auctions
Rep
Auctioneer
Desired Tasks
Desired Processes
Executor
CellGarden
Containers
Bids
RepExecuto
r
CellGarden
Containers
Garden Windows
Garden Linux
Garden API
Garden Container AbstractionA platform-neutral API for containerization
Run Dockerized applications
Run .NET applications
Run workers and tasks
Develop Cloud Foundry applications locally
Extending the Cloud Foundry Runtime
Extend the deployment API to run a Docker image on our native Garden-Linux container
backend. cf push APP DOCKER-REPO
POST /containers{ "bind_mounts": [], "grace_time": 1200, "handle": 'user-supplied-handle', "network": 'network', "rootfs": 'rootfs', "properties": [], "env": [] }
200 Ok{ handle: 'handle-of-created-container' }
Deploy the container image root filesystem.
Attempt to parse container image metadata for memory, disk, user, launch command and port.
Container Scheduler &Health Monitor
Windows CellsContainer Execution Nodes
Shared Runtime State
Service Discovery & Monitoring
New Workload Types
Tasks
A single unit of work
Runs at most once
N long running instances
Distributed across cells for HA
Monitored and restarted
Long Running
Processes
RunAction: run process in container
DownloadAction: fetches and extract archive
UploadAction: POST file from container to URL
ParallelAction: run multiple actions in parallel
SerialAction: runs multiple actions in order
EmitProgressAction: wraps action and logs
progress
TimeoutAction: wrap action and fail if timed
out
TryAction: wrap action and ignore errors
WorkloadPrimitives
Desired Task
{ ...
"rootfs": "docker:///docker-org/docker-image", "env": [ {"name": "ENV_NAME_A", "value": "ENV_VALUE_A"}, {"name": "ENV_NAME_B", "value": "ENV_VALUE_B"} ],
"cpu_weight": 57, "disk_mb": 1024, "memory_mb": 128, "privileged": true,
"action": ACTION(s) TO RUN,
...}
Desired LRP
{ ... "instances": 17, "rootfs": "VALID-ROOTFS", "env": [ {"name": "ENV_NAME_A", "value": "ENV_VALUE_A"}, {"name": "ENV_NAME_B", "value": "ENV_VALUE_B"} ], "cpu_weight": 57, "disk_mb": 1024, "memory_mb": 128, "privileged": true, "setup": ACTION, "action": ACTION, "monitor": ACTION, "start_timeout": N seconds, "ports": [8080, 5050], ...}
• Single-tenant• Everyone is ‘cluster root’• Wide-open networking• Not all components are HA• Red-black upgrades• No data services
• Multi-tenancy with resource quotas• Role-based access control• Application security groups• Highly-available components• Zero-downtime, rolling upgrades• Backing data service orchestration
Repackaging the Runtime
Production usage with 20+ VMs
Local development on a single host
10,000 “real app” container instances (100 per cell)
4,000 concurrent tasks4,000-instance LRPs
ScalabilityRuntime testing with tens of thousands of containers: