Using Docker in Cloud Foundry
This topic describes how Cloud Foundry (CF) operators can enable CF developers to run their apps in Docker containers, and explains how Docker works in Cloud Foundry.
For information about Diego, the Cloud Foundry component that manages application containers, see the Diego Architecture topic. For information about how CF developers push apps with Docker images, see Deploying an App with Docker.
By default, apps deployed with the
cf push command run in standard Cloud Foundry Linux containers. With Docker support enabled, Cloud Foundry can also deploy and manage apps running in Docker containers.
To deploy apps to Docker, developers run
cf push with the
--docker-image option and the location of a Docker image to create the containers from. See the Push a Docker Image topic for information about how CF developers push apps with Docker images.
To enable Docker support on a CF deployment, an operator must do the following:
Configure access to any Docker registries that developers want to use images from.
diego_docker feature flag governs whether a CF deployment supports Docker containers.
To enable Docker support, run:
$ cf enable-feature-flag diego_docker
To disable Docker support, run:
$ cf disable-feature-flag diego_docker
Note: Disabling the
diego_docker feature flag stops all Docker-based apps in your deployment within a few convergence cycles, on the order of a minute.
To support Docker, Pivotal Cloud Foundry needs the ability to access Docker registries using either a Certificate Authority or an IP address whitelist. The Using Docker Registries topic explains how to configure this access.
A Docker image consists of a collection of layers. Each layer consists of one or both of the following:
- Raw bits to download and mount. These bits form the file system.
- Metadata that describes commands, users, and environment for the layer.
This metadata includes the
CMDdirectives, and is specified in the Dockerfile.
Diego currently uses Garden-runC to construct Linux containers.
Both Docker and Garden-runC use libraries from the Open Container Initiative (OCI) to build Linux containers. After creation, these containers use name space isolation, or namespaces, and control groups, or cgroups, to isolate processes in containers and limit resource usage. These are common kernel resource isolation features used by all Linux containers.
Note: PAS versions v1.8.8 and above use Garden-runC instead of Garden-Linux.
Before Garden-runC creates a Linux container, it creates a file system that is mounted as the root file system of the container. Garden-runC supports mounting Docker images as the root file systems for the containers it creates.
When creating a container, both Docker and Garden-runC perform the following actions:
- Fetch and cache the individual layers associated with a Docker image
- Combine and mount the layers as the root file system
These actions produce a container whose contents exactly match the contents of the associated Docker image.
Earlier versions of Diego used Garden-Linux. For more information, see the Garden topic.
After Garden-runC creates a container, Diego runs and monitors the processes inside of it.
To determine which processes to run, the Cloud Controller fetches and stores the metadata associated with the Docker image. The Cloud Controller uses this metadata to perform the following actions:
- Runs the start command as the user specified in the Docker image
- Instructs Diego and the Gorouter to route traffic to the lowest-numbered port exposed in the Docker image, or port 8080 if the Dockerfile does not explicitly expose a listen port.
Note: When launching an application on Diego, the Cloud Controller honors any user-specified overrides such as a custom start command or custom environment variables.
The attack surface area for a Docker-based container running on Diego remains somewhat higher than that of a buildpack application because Docker allows users to fully specify the contents of their root file systems. A buildpack application runs on a trusted root filesystem.
Garden-runC provides features that allow the platform to run Docker images more
securely in a multi-tenant context.
In particular, Cloud Foundry uses the
user-namespacing feature found on
modern Linux kernels to ensure that users cannot gain escalated privileges on
the host even if they escalate privileges within a container.
The Cloud Controller always runs Docker containers on Diego with user namespaces enabled. This security restriction prevents certain features, such as the ability to mount FuseFS devices, from working in Docker containers. Docker applications can use fuse mounts through volume services, but they cannot directly mount fuse devices from within the container.
To mitigate security concerns, Cloud Foundry recommends that you run only trusted Docker containers on the platform. By default, the Cloud Controller does not allow Docker-based applications to run on the platform.