spec Documentation

1. Project Overview & Quickstart (devcontainers/spec) Copy Chapter

File: README.md Copy Chapter

Development Containers

This repository is for the Development Container Specification. You can check out the spec on our website https://containers.dev/ too.

A development container allows you to use a container as a full-featured development environment. It can be used to run an application, to separate tools, libraries, or runtimes needed for working with a codebase, and to aid in continuous integration and testing.

The Development Container Specification seeks to find ways to enrich existing formats with common development specific settings, tools, and configuration while still providing a simplified, un-orchestrated single container option – so that they can be used as coding environments or for continuous integration and testing.

The first format in the specification, devcontainer.json, was born out of necessity. It is a structured JSON with Comments (jsonc) metadata format that tools can use to store any needed configuration required to develop inside of local or cloud-based containerized coding.

We envision that this same structured data can be embedded in images and other formats – all while retaining a common object model for consistent processing. For example, some of this same metadata can be added to a devcontainer.metadata image label to tie settings directly to a container image.

Beyond repeatable setup, these same development containers provide consistency to avoid environment specific problems across developers and centralized build and test automation services. You can use the open-source CLI reference implementation either directly or integrated into product experiences to use the structured metadata to deliver these benefits. It currently supports integrating with Docker Compose and a simplified, un-orchestrated single container option – so that they can be used as coding environments or for continuous integration and testing.

A GitHub Action and an Azure DevOps Task are available in devcontainers/ci for running a repository's dev container in continuous integration (CI) builds. This allows you to reuse the same setup that you are using for local development to also build and test your code in CI.

Spec content Copy Section

You may review the specification in the docs/specs folder of this repo.

You may also review active proposals in the proposals folder.

Images used in this repo will be contained in the images folder. The icon for the dev container GitHub org is from the Fluent icon library.

Contributing and Feedback Copy Chapter

If you are interested in contributing, please check out the How to Contribute document, open an issue, or join our community Slack channel.

Please report issues in the following repositories:

• Spec-maintained Features and Templates: devcontainers/features, devcontainers/templates
• CLI reference implementation and non-spec related feature requests: devcontainers/cli
• GitHub Action and Azure DevOps Task: devcontainers/ci

License

License for this repository:

Copyright © Microsoft Corporation All rights reserved.
Creative Commons Attribution 4.0 License (International): https://creativecommons.org/licenses/by/4.0/legalcode

File: docs/specs/declarative-secrets.md Copy Chapter

Declarative Secrets

This has now been implemented:

• Discussion issue: https://github.com/devcontainers/spec/issues/198, https://github.com/devcontainers/spec/issues/216
• Schema PR: https://github.com/devcontainers/spec/pull/303

Below is the original proposal.

Motivation Copy Chapter

Various projects exist in the wild that require various secrets for them to run properly. Examples include:

• https://github.com/lostintangent/codespaces-langchain
• https://github.com/openai/openai-quickstart-python
• https://github.com/openai/openai-quickstart-node

Today these projects have to include specific instructions (e.g. in their README) telling users where to procure what they need to run and then how to set it up as a secret or add it as an environment variable. This currently acts as an impediment to adoption and promotion of dev containers for these projects.

Goal Copy Chapter

Simplify using dev containers for these kinds of projects by supporting secrets as a first-class part of the dev container creation flow.

Proposal Copy Chapter

Add an optional secrets property to the devContainer.base.schema.json. This will be used to declare the secrets needed within the dev container.

The secrets property contains a map of secret names and details about those secrets. These secrets are recommended to the user but are not required for creation. All properties are optional.

Example Copy Chapter

{
    "image": "mcr.microsoft.com/devcontainers/base:bullseye",
    "secrets": {
        "CHAT_GPT_API_KEY": {
            "description": "I'm your super cool API key for ChatGPT.",
            "documentationUrl": "https://openai.com/api/"
        },
        "STABLE_DIFFUSION_API_KEY": {}
    }
}

This example would declare that this dev container wants the user to provide two secrets. The CHAT_GPT_API_KEY secret also provides some additional metadata that clients can use in displaying a user experience to provide that secret. The STABLE_DIFFUSION_API_KEY skips that metadata.

Implementations may inject these secrets into the container as environment variables.

Possible Future Extensions Copy Chapter

These are out of scope for this proposal but the following has been discussed in relation:

1. Inclusion of a type field per secret. The default, which we are assuming and leveraging here, would be "environmentVariable". This would include an option to specify "reference" and then reference the provided secrets in other parts of the dev container.
2. Prescriptions of how the secrets are injected into the dev container.

File: docs/specs/devcontainer-features.md Copy Chapter

Dev Container Features reference

Development Container Features are self-contained, shareable units of installation code and development container configuration. The name comes from the idea that referencing one of them allows you to quickly and easily add more tooling, runtime, or library "Features" into your development container for you or your collaborators to use.

Note: While Features may be installed on top of any base image, the implementation of a Feature might restrict it to a subset of possible base images.

For example, some Features may be authored to work with a certain linux distro (e.g. debian-based images that use the apt package manager).

Feature metadata is captured by a devcontainer-feature.json file in the root folder of the Feature.

Tip: This section covers details on the Features specification. If you are looking for summarized information on creating your own Features, see the template and core Features repositories.

Folder structure Copy Chapter

A Feature is a self contained entity in a folder with at least a devcontainer-feature.json and install.sh entrypoint script. Additional files are permitted and are packaged along side the required files.

+-- feature
|    +-- devcontainer-feature.json
|    +-- install.sh
|    +-- (other files)

`devcontainer-feature.json` properties Copy Chapter

The devcontainer-feature.json file defines metadata about a given Feature.

All properties are optional except for id, version, and name.

(**) The ID must refer to either a Feature (1) published to an OCI registry, (2) a Feature Tgz URI, or (3) a Feature in the local file tree. Deprecated Feature identifiers (i.e. GitHub Release) are not supported and the presence of this property may be considered a fatal error or ignored. For local Features (i.e. during development), you may also depend on other local Features by providing a relative path to the Feature, relative to folder containing the active devcontainer.json. This behavior of Features within this property again mirror the features object in devcontainer.json.

Lifecycle Hooks Copy Section

The following lifecycle hooks may be declared as properties of devcontainer-feature.json.

Behavior

Each property mirrors the behavior of the matching property in devcontainer.json, including the behavior that commands are executed from the context of the project workspace folder.

For each lifecycle hook (in Feature installation order), each command contributed by a Feature is executed in sequence (blocking the next command from executing). Commands provided by Features are always executed before any user-provided lifecycle commands (i.e. in the devcontainer.json).

If a Feature provides a given command with the object syntax, all commands within that group are executed in parallel, but still blocking commands from subsequent Features and/or the devcontainer.json.

Note: These properties are stored within image metadata.

The `options` property Copy Section

The options property is contains a map of option IDs and their related configuration settings. The ID becomes the name of the environment variable in all caps. See option resolution for more details. For example:

{
  "options": {
    "optionIdGoesHere": {
      "type": "string",
      "description": "Description of the option",
      "proposals": ["value1", "value2"],
      "default": "value1"
    }
  }
}

User environment variables Copy Section

Feature scripts run as the root user and sometimes need to know which user account the dev container will be used with.

_REMOTE_USER and _CONTAINER_USER environment variables are passed to the Features scripts with _CONTAINER_USER being the container's user and _REMOTE_USER being the configured remoteUser. If no remoteUser is configured, _REMOTE_USER is set to the same value as _CONTAINER_USER.

Additionally, the home folders of the two users are passed to the Feature scripts as _REMOTE_USER_HOME and _CONTAINER_USER_HOME environment variables.

The container user can be set with containerUser in the devcontainer.json and image metadata, user in the docker-compose.yml, USER in the Dockerfile, and can be passed down from the base image.

Dev Container ID Copy Section

An identifier will be referred to as ${devcontainerId} in the devcontainer.json and the Feature metadata and that will be replaced with the dev container's id. It should only be used in parts of the configuration and metadata that is not used for building the image because that would otherwise prevent pre-building the image at a time when the dev container's id is not known yet. Excluding boolean, numbers and enum properties the properties supporting ${devcontainerId} in the Feature metadata are: entrypoint, mounts, customizations.

Implementations can choose how to compute this identifier. They must ensure that it is unique among other dev containers on the same Docker host and that it is stable across rebuilds of dev containers. The identifier must only contain alphanumeric characters. We describe a way to do this below.

Label-based Implementation

The following assumes that a dev container can be identified among other dev containers on the same Docker host by a set of labels on the container. Implementations may choose to follow this approach.

The identifier is derived from the set of container labels uniquely identifying the dev container. It is up to the implementation to choose these labels. For example, if the dev container is based on a local folder, the label could be named devcontainer.local_folder and have the local folder's path as its value.

For example, the ghcr.io/devcontainers/features/docker-in-docker Feature could use the dev container id with:

{
    "id": "docker-in-docker",
    "version": "1.0.4",
    // ...
    "mounts": [
        {
            "source": "dind-var-lib-docker-${devcontainerId}",
            "target": "/var/lib/docker",
            "type": "volume"
        }
    ]
}

Label-based Computation

• Input the labels as a JSON object with the object's keys being the label names and the object's values being the labels' values.
  • To ensure implementations get to the same result, the object keys must be sorted and any optional whitespace outside of the keys and values must be removed.
• Compute a SHA-256 hash from the UTF-8 encoded input string.
• Use a base-32 encoded representation left-padded with '0' to 52 characters as the result.

JavaScript implementation taking an object with the labels as argument and returning a string as the result:

const crypto = require('crypto');
function uniqueIdForLabels(idLabels) {
	const stringInput = JSON.stringify(idLabels, Object.keys(idLabels).sort()); // sort properties
	const bufferInput = Buffer.from(stringInput, 'utf-8');
	const hash = crypto.createHash('sha256')
		.update(bufferInput)
		.digest();
	const uniqueId = BigInt(`0x${hash.toString('hex')}`)
		.toString(32)
		.padStart(52, '0');
	return uniqueId;
}

devcontainer.json properties Copy Chapter

Features are referenced in a user's devcontainer.json under the top level features object.

A user can specify an arbitrary number of Features. At build time, these Features will be installed in an order defined by a combination of the installation order rules and implementation.

A single Feature is provided as a key/value pair, where the key is the Feature identifier, and the value is an object containing "options" (or empty for "default"). Each key in the Feature object must be unique.

These options are sourced as environment variables at build-time, as specified in Option Resolution.

Below is a valid features object provided as an example.

"features": {
  "ghcr.io/user/repo/go": {},
  "ghcr.io/user/repo1/go:1": {},
  "ghcr.io/user/repo2/go:latest": {},
  "https://github.com/user/repo/releases/devcontainer-feature-go.tgz": { 
        "optionA": "value" 
  },
  "./myGoFeature": { 
        "optionA": true,
        "optionB": "hello",
        "version" : "1.0.0"
  }
}

Note: The :latest version annotation is added implicitly if omitted. To pin to a specific package version (example), append it to the end of the Feature.

An option's value can be provided as either a string or boolean, and should match what is expected by the Feature in the devcontainer-feature.json file.

As a shorthand, the value of a feature can be provided as a single string. This string is mapped to an option called version. In the example below, both examples are equivalent.

"features": {
  "ghcr.io/owner/repo/go": "1.18"
}

"features": {
  "ghcr.io/owner/repo/go": {
    "version": "1.18"
  }
}

Referencing a feature Copy Section

The id format specified dictates how a supporting tool will locate and download a given Feature. id is one of the following:

Feature identifiers should be assumed to be case-insensitive and should be normalized to lowercase.

(*) OCI registry must implement the OCI Artifact Distribution Specification. Some implementors can be found here.

(**) The provided path is always relative to the folder containing the devcontainer.json. Further requirements are outlined in the Locally Referenced Addendum

Versioning Copy Chapter

Each Feature is individually versioned according to the semver specification. The version property in the respective devcontainer-feature.json file is updated to increment the Feature's version.

Tooling that handles releasing Features will not republish Features if that exact version has already been published; however, tooling must republish major and minor versions in accordance with the semver specification.

Authoring Copy Chapter

Features can be authored in a number of languages, the most straightforward being bash scripts. If a Feature is authored in a different language information about it should be included in the metadata so that users can make an informed choice about it.

Reference information about the application required to execute the Feature should be included in devcontainer-feature.json in the metadata section.

Applications should default to /bin/sh for Features that do not include this information.

If the Feature is included in a folder as part of the repository that contains devcontainer.json, no other steps are necessary.

Release Copy Chapter

For information on distributing Features, see devcontainer-features-distribution.md.

Execution Copy Chapter

Invoking `install.sh` Copy Section

The install.sh script for each Feature should be executed as root during a container image build. This allows the script to add needed OS dependencies or settings that could not otherwise be modified. This also allows the script to switch into another user's context using the su command (e.g. su ${USERNAME} -c "command-goes-here"). In combination, this allows both root and non-root image modifications to occur even if sudo is not present in the base image for security reasons.

To ensure that the appropriate shell is used, the execute bit should be set on install.sh and the file invoked directly (e.g. chmod +x install.sh && ./install.sh).

Note: It is recommended that Feature authors write install.sh using a shell available by default in their supported distributions (e.g. bash in Debian/Ubuntu or Fedora, sh in Alpine). In the event a different shell is required (e.g. fish), install.sh can be used to boostrap by checking for the presence of the desired shell, installing it if needed, and then invoking a secondary script using the shell.

The install.sh file can similarly be used to bootstrap something written in a compiled language like Go. Given the increasing likelihood that a Feature needs to work on both x86_64 and arm64-based devices (e.g. Apple Silicon Macs), install.sh can detect the current architecture (e.g. using something like uname -m or dpkg --print-architecture), and then invoke the right executable for that architecture.

Installation order Copy Section

By default, Features are installed on top of a base image in an order determined as optimal by the implementing tool.

If any of the following properties are provided in the Feature's devcontainer-feature.json, or the user's devcontainer.json, the order indicated by these propert(ies) are respected.

• The dependsOn property defined as a part of a Feature's devcontainer-feature.json.
• The installsAfter property defined as part of a Feature's devcontainer-feature.json.
• The overrideFeatureInstallOrder property in user's devcontainer.json. Allows users to control the order of execution of their Features.

The `dependsOn` property

The optional dependsOn property indicates a set of required, "hard" dependencies for a given Feature.

The dependsOn property is declared in a Feature's devcontainer-feature.json metadata file. Elements of this property mirror the semantics of the features object in devcontainer.json. Therefore, all dependencies may provide the relevant options, or an empty object (e.g. "bar:123": {}) if the Feature's default options are sufficient. Identical Features that provide different options are treated as different Features (see Feature equality for more info).

All Features indicated in the dependsOn property must be satisfied (a Feature equal to each dependency is present in the installation order) before the given Feature is set to be installed. If any of the Features indicated in the dependsOn property cannot be installed (e.g. due to circular dependency, failure to resolve the Feature, etc) the entire dev container creation should fail.

The dependsOn property must be evaluated recursively. Therefore, if a Feature dependency has its own dependsOn property, that Feature's dependencies must also be satisfied before the given Feature is installed.

{
    "name": "My Feature",
    "id": "myFeature",
    "version": "1.0.0",
    "dependsOn": {
        "foo:1": {
            "flag": true
        },
        "bar:1.2.3": {},
        "baz@sha256:a4cdc44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" {},
    }
}

In the snippet above, myfeature MUST be installed after foo, bar, and baz. If the Features provided via the dependsOn property declare their own dependencies, those must also be satisfied before the Feature is installed.

The `installsAfter` Property

The installsAfter property indicates a "soft dependency" that influences the installation order of Features that are already queued to be installed. The effective behavior of this property is the same as dependsOn, with the following differences:

• installsAfter is not evaluated recursively.
• installsAfter only influences the installation order of Features that are already set to be installed. Any Feature not set to be installed after (1) resolving the dependsOn dependency tree or (2) indicated by the user's devcontainer.json should not be added to the installation list.
• The Feature indicated by installsAfter can not provide options, nor are they able to be pinned to a specific version tag or digest. Resolution to the canonical name should still be performed (e.g. if the Feature has been renamed).

{
    "name": "My Feature",
    "id": "myFeature",
    "version": "1.0.0",
    "installsAfter": [
        "foo",
        "bar"
    ]
}

In the snippet above, myfeature must be installed after foo and bar if the Feature is already queued to be installed. If foo and bar are not already queued to be installed, this dependency relationship should be ignored.

The 'overrideFeatureInstallOrder' property

The overrideFeatureInstallOrder property of devcontainer.json is an array of Feature IDs that are to be installed in descending priority order as soon as its dependencies outlined above are installed.

This property may not indicate an installation order that is inconsistent with the resolved dependency graph (see dependency algorithm). If the overrideFeatureInstallOrder property is inconsistent with the dependency graph, the implementing tool should fail the dependency resolution step.

This evaluation is performed by assigning a roundPriority to all nodes that match the Feature identifier (version omitted) present in the property.

For example, given n Features in the overrideFeatureInstallOrder array, the orchestrating tool should assign a roundPriority of n - idx to each Feature, where idx is the zero-based index of the Feature in the array.

For example:

overrideFeatureInstallOrder = [
  "foo",
  "bar",
  "baz"
]

would result in the following roundPriority assignments:

const roundPriority = {
  "foo": 3,
  "bar": 2,
  "baz": 1
}

This property must not influence the dependency relationship as defined by the dependency graph (see dependency graph) and shall only be evaulated at the round-based sorting step (see round sort). Put another way, this property cannot "pull forward" a Feature until all of its dependencies (both soft and hard) have been installed. After a Feature's dependencies have been installed in other rounds, this property should "pull forward" each Feature as early as possible (given the order of identifiers in the array).

Similar to installsAfter, this property's members may not provide options, nor are they able to be pinned to a specific version tag or digest.

If a Feature is indicated in overrideFeatureInstallOrder but not a member of the dependency graph (i.e. it is not queued to be installed), the orchestrating tool may fail the dependency resolution step.

Definitions Copy Chapter

Definition: Feature Equality Copy Section

This specification defines two Features as equal if both Features point to the same exact contents and are executed with the same options.

For Features published to an OCI registry, two Feature are identical if their manifest digests are equal, and the options executed against the Feature are equal (compared value by value). Identical manifest digests implies that the tgz contents of the Feature and its entire devcontainer-feature.json are identical. If any of these conditions are not met, the Features are considered not equal.

For Features fetched by HTTPS URI, two Features are identical if the contents of the tgz are identical (hash to the same value), and the options executed against the Feature are equal (compared value by value). If any of these conditions are not met, the Features are considered not equal.

For local Features, each Feature is considered unique and not equal to any other local Feature.

Definition: Round Stable Sort Copy Section

To prevent non-deterministic behavior, the algorithm will sort each round according to the following rules:

• Compare and sort each Feature lexiographically by their fully qualified resource name (For OCI-published Features, that means the ID without version or digest.). If the comparison is equal:
• Compare and sort each Feature from oldest to newest tag (latest being the "most new"). If the comparision is equal:
• Compare and sort each Feature by their options by:
  • Greatest number of user-defined options (note omitting an option will default that value to the Feature's default value and is not considered a user-defined option). If the comparison is equal:
  • Sort the provided option keys lexicographically. If the comparison is equal:
  • Sort the provided option values lexicographically. If the comparision is equal:
• Sort Features by their canonical name (For OCI-published Features, the Feature ID resolved to the digest hash).

If there is no difference based on these comparator rules, the Features are considered equal.

Dependency installation order algorithm Copy Chapter

An implementing tool is responsible for calculating the Feature installation order (or providing an error if no valid installation order can be resolved). The set of Features to be installed is the union of user-defined Features (those directly indicated in the user's devcontainer.json and their dependencies (those indicated by the dependsOn or installsAfter property, taking into account the user dev container's overrideFeatureInstallOrder property). The implmenting tool will perform the following steps:

(1) Build a dependency graph Copy Section

From the user-defined Features, the orchestrating tool will build a dependency graph. The graph will be built by traversing the dependsOn and installsAfter properties of each Feature. The metadata for each dependency is then fetched and the node added as an edge to to the dependent Feature. For dependsOn dependencies, the dependency will be fed back into the worklist to be recursively resolved.

An accumulator is maintained with all uniquely discovered and user-provided Features, each with a reference to its dependencies. If the exact Feature (see Feature Equality) has already been added to the accumulator, it will not be added again. The accumulator will be fed into (B3) after the Feature tree has been resolved.

The graph may be stored as an adjacency list with two kinds of edges (1) dependsOn edges or "hard dependencies" and (2) installsAfter edges or "soft dependencies".

(2) Assigning round priority Copy Section

Each node in the graph has an implicit, default roundPriority of 0.

To influence installation order globally while still honoring the dependency graph of built in (1), roundPriority values may be tweaks for each Feature. When each round is calculated in (3), only the Features equal to the max roundPriority of that set will be committed (the remaining will be uncommitted and reevaulated in subsequent rounds).

The roundPriority is set to a non-zero value in the following instances:

• If the devcontainer.json contains an overrideFeatureInstallOrder.

(3) Round-based sorting Copy Section

Perform a sort on the result of (1) in rounds. This sort will rearrange Features, producing a sorted list of Features to install. The sort will be performed as follows:

Start with all the elements from (2) in a worklist and an empty list installationOrder. While the worklist is not empty, iterate through each element in the worklist and check if all its dependencies (if any) are already members of installationOrder. If the check is true, add it to an intermediate list round If not, skip it. Equality is determined in Feature Equality.

Then for each intermediate round list, commit to installationOrder only those nodes who share the maximum roundPriority. Return all nodes in round with a strictly lower roundPriority to the worklist to be reprocessed in subsequent iterations. If there are multiple nodes with the same roundPriority, commit them to installationOrder with a final sort according to Round Stable Sort.

Repeat for as many rounds as necessary until worklist is empty. If there is ever a round where no elements are added to installationOrder, the algorithm should terminate and return an error. This indicates a circular dependency or other fatal error in the dependency graph. Implementations should attempt to provide the user with information about the error and possible mitigation strategies.

Notes Copy Section

From an implementation point of view, installsAfter nodes may be added as a separate set of directed edges, just as dependsOn nodes are added as directed edges (see (1)). Before round-based installation and sorting (3), an orchestrating tool should remove all installsAfter directed edges that do not correspond with a Feature in the worklist that is set to be installed. In each round, a Feature can then be installed if all its requirements (both dependsOn and installsAfter dependencies) have been fulfilled in previous rounds.

An implemention should fail the dependency resolution step if the evaluation of the installsAfter property results in an inconsistent state (e.g. a circular dependency).

Option Resolution Copy Section

A Feature's 'options' - specified as the value of a single Feature key/value pair in the user's devcontainer.json - are passed to the Feature as environment variables.

A supporting tool will parse the options object provided by the user. If a value is provided for a Feature, it will be emitted to a file named devcontainer-features.env following the format <OPTION_NAME>=<value>.

To ensure a option that is valid as an environment variable, the follow substitutions are performed.

(str: string) => str
	.replace(/[^\w_]/g, '_')
	.replace(/^[\d_]+/g, '_')
	.toUpperCase();

This file is sourced at build-time for the Feature install.sh entrypoint script to handle.

Any options defined by a Feature's devcontainer-feature.json that are omitted in the user's devcontainer.json will be implicitly exported as its default value.

Option resolution example Copy Section

Suppose a python Feature has the following options parameters declared in the devcontainer-feature.json file:

// ...
"options": {
    "version": {
        "type": "string",
        "enum": ["latest", "3.10", "3.9", "3.8", "3.7", "3.6"],
        "default": "latest",
        "description": "Select a Python version to install."
    },
    "pip": {
        "type": "boolean",
        "default": true,
        "description": "Installs pip"
    },
    "optimize": {
        "type": "boolean",
        "default": true,
        "description": "Optimize python installation"
    }
}

The user's devcontainer.json declared the python Feature like so

"features": {
    "ghcr.io/devcontainers/features/python:1": {
        "version": "3.10",
        "pip": false
    }
}

The emitted environment variables will be:

VERSION="3.10"
PIP="false"
OPTIMIZE="true"

These will be sourced and visible to the install.sh entrypoint script. The following install.sh...

#!/usr/bin/env bash

echo "Version is $VERSION"
echo "Pip? $PIP"
echo "Optimize? $OPTIMIZE"

... outputs the following:

Version is 3.10
Pip? false
Optimize? true

Steps to rename a Feature Copy Section

1. Update the Feature source code folder and the id property in the devcontainer-feature.json properties to reflect the new id. Other properties (name, documentationUrl, etc.) can optionally be updated during this step.
2. Add or update the legacyIds property to the Feature, including the previously used id.
3. Bump the semantic version of the Feature.
4. Rerun the devcontainer features publish command, or equivalent tool that implements the Features distribution specification.

Example: Renaming a Feature

Let's say we currently have a docker-from-docker Feature 👇

Current devcontainer-feature.json :

{
    "id": "docker-from-docker",
    "version": "2.0.1",
    "name": "Docker (Docker-from-Docker)",
    "documentationURL": "https://github.com/devcontainers/features/tree/main/src/docker-from-docker",
    ....
}

We'd want to rename this Feature to docker-outside-of-docker. The source code folder of the Feature will be updated to docker-outside-of-docker and the updated devcontainer-feature.json will look like 👇

{
    "id": "docker-outside-of-docker",
    "version": "2.0.2",
    "name": "Docker (Docker-outside-of-Docker)",
    "documentationURL": "https://github.com/devcontainers/features/tree/main/src/docker-outside-of-docker",
    "legacyIds": [
        "docker-from-docker"
    ]
    ....
}

Note - The semantic version of the Feature defined by the version property should be continued and should not be restarted at 1.0.0.

Implementation notes Copy Section

There are several things to keep in mind for an application that implements Features:

• The order of execution of Features is determined by the application, based on the installsAfter property used by feature authors. It can be overridden by users if necessary with the overrideFeatureInstallOrder in devcontainer.json.
• Features are used to create an image that can be used to create a container or not.
• Parameters like privileged, init are included if just 1 Feature requires them.
• Parameters like capAdd, securityOp are concatenated.
• containerEnv is added before the Feature is executed as ENV commands in the Dockerfile.
• Each Feature script executes as its own layer to aid in caching and rebuilding.

File: docs/specs/devcontainer-features-distribution.md Copy Chapter

Distribution and Discovery of Dev Container Features

TL;DR Check out the quick start repository to get started on distributing your own Dev Container Features.

This specification defines a pattern where community members and organizations can author and self-publish Dev Container Features.

Goals include:

• For Feature authors, create a "self-service" way to publish a Feature, either publicly or privately, that is not centrally controlled.
• For users, provide the ability to validate the integrity of fetched Feature assets.
• For users, provide the ability to pin to a particular version (absolute, or semantic version) of a Feature to allow for consistent, repeatable environments.
• Provide the ability to standardize publishing such that supporting tools may implement their own mechanism to aid Feature discoverability as they see fit.

Tip: This section covers details on the Features specification. If you are looking for summarized information on creating your own Features, see the template and core Features repositories.

Source code Copy Chapter

Features source code is stored in a git repository.

For ease of authorship and maintenance, [1..n] features can share a single git repository. This set of features is referred to as a "collection," and will share the same devcontainer-collection.json file and "namespace" (eg. <owner>/<repo>).

Source code for the set follows the example file structure below:

.
├── README.md
├── src
│   ├── dotnet
│   │   ├── devcontainer-feature.json
│   │   ├── install.sh
│   │   └── ...
|   ├
│   ├── go
│   │   ├── devcontainer-feature.json
│   │   └── install.sh
|   ├── ...
│   │   ├── devcontainer-feature.json
│   │   └── install.sh
├── test
│   ├── dotnet
│   │   ├── test.sh
│   │   └── ...
│   └── go
│   |   └── test.sh
|   ├── ...
│   │   └── test.sh
├── ...

...where src is a directory containing a sub-folder with the name of the Feature (e.g. src/dotnet or src/go) with at least a file named devcontainer-feature.json that contains the Feature metadata, and an install.sh script that implementing tools will use as the entrypoint to install the Feature.

Each sub-directory should be named such that it matches the id field of the devcontainer-feature.json. Other files can also be included in the Feature's sub-directory, and will be included during the packaging step alongside the two required files. Any files that are not part of the Feature's sub-directory (e.g. outside of src/dotnet) will not be included in the packaging step.

Optionally, a mirrored test directory can be included with an accompanying test.sh script. Implementing tools may use this to run tests against the given Feature.

Versioning Copy Chapter

Each Feature is individually versioned according to the semver specification. The version property in the respective devcontainer-feature.json file is parsed to determine if the Feature should be republished.

Tooling that handles publishing Features will not republish features if that exact version has already been published; however, tooling must republish major and minor versions in accordance with the semver specification.

Packaging Copy Chapter

Features are distributed as tarballs. The tarball contains the entire contents of the Feature sub-directory, including the devcontainer-feature.json, install.sh, and any other files in the directory.

The tarball is named devcontainer-feature-.tgz, where `` is the Feature's id field.

A reference implementation for packaging and distributing Features is provided as a GitHub Action (https://github.com/devcontainers/action).

devcontainer-collection.json Copy Section

The devcontainer-collection.json is an auto-generated metadata file.

Each Features's devcontainer-feature.json metadata file is appended into the features top-level array.

Distribution Copy Chapter

There are several supported ways to distribute Features. Distribution is handled by the implementing packaging tool such as the Dev Container CLI or Dev Container Publish GitHub Action. See the quick start repository for a full working example.

Feature identifiers should be provided lowercase by an author. If not, an implementing packaging tool should normalize the identifier into lowercase for any internal representation.

A user references a distributed Feature in a devcontainer.json as defined in 'referencing a feature'.

OCI Registry Copy Section

An OCI registry that implements the OCI Artifact Distribution Specification serves as the primary distribution mechanism for Features.

Each packaged feature is pushed to the registry following the naming convention <registry>/<namespace>/[:version], where version is the major, minor, and patch version of the Feature, according to the semver specification.

Note: namespace is a unique identifier for the collection of Features. There are no strict rules for namespace; however, one pattern is to set namespace equal to source repository's <owner>/<repo>. The namespace should be lowercase, following the regex provided in the OCI specification.

A custom media type application/vnd.devcontainers and application/vnd.devcontainers.layer.v1+tar are used as demonstrated below.

For example, the go Feature in the devcontainers/features namespace at version 1.2.3 would be pushed to the ghcr.io OCI registry.

Note: The example below uses oras for demonstration purposes. A supporting tool should directly implement the required functionality from the aforementioned OCI artifact distribution specification.

# ghcr.io/devcontainers/features/go:1 
REGISTRY=ghcr.io
NAMESPACE=devcontainers/features
FEATURE=go

ARTIFACT_PATH=devcontainer-feature-go.tgz

for VERSION in 1  1.2  1.2.3  latest
do
    oras push ${REGISTRY}/${NAMESPACE}/${FEATURE}:${VERSION} \
            --manifest-config /dev/null:application/vnd.devcontainers \
                             ./${ARTIFACT_PATH}:application/vnd.devcontainers.layer.v1+tar
done

The "namespace" is the globally identifiable name for the collection of Features (e.g. owner/repo for the source code's git repository).

The auto-generated devcontainer-collection.json is pushed to the registry with the same namespace as above and no accompanying feature name. The collection file is always tagged as latest.

# ghcr.io/devcontainers/features
REGISTRY=ghcr.io
NAMESPACE=devcontainers/features

oras push ${REGISTRY}/${NAMESPACE}:latest \
        --manifest-config /dev/null:application/vnd.devcontainers \
                            ./devcontainer-collection.json:application/vnd.devcontainers.collection.layer.v1+json

Additionally, an annotation named dev.containers.metadata should be populated on the manifest when published by an implementing tool. This annotation is the escaped JSON object of the entire devcontainer-feature.json as it appears during the packaging stage.

An example manifest with the dev.containers.metadata annotation:

{
  "schemaVersion": 2,
  "mediaType": "application/vnd.oci.image.manifest.v1+json",
  "config": {
    "mediaType": "application/vnd.devcontainers",
    "digest": "sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
    "size": 0
  },
  "layers": [
    {
      "mediaType": "application/vnd.devcontainers.layer.v1+tar",
      "digest": "sha256:738af5504b253dc6de51d2cb1556cdb7ce70ab18b2f32b0c2f12650ed6d2e4bc",
      "size": 3584,
      "annotations": {
        "org.opencontainers.image.title": "devcontainer-feature-myFeature.tgz"
      }
    }
  ],
  "annotations": {
    "dev.containers.metadata": "{\"name\": \"My Feature\",\"id\": \"myFeature\",\"version\": \"1.0.0\",\"dependsOn\": {\"ghcr.io/myotherFeature:1\": {\"flag\": true},\"features.azurecr.io/aThirdFeature:1\": {},\"features.azurecr.io/aFourthFeature:1.2.3\": {}}}"
  }
}

Directly referencing a tarball Copy Section

A Feature can be referenced directly in a user's devcontainer.json file by HTTPS URI that points to the tarball from the package step.

The .tgz archive file must be named devcontainer-feature-<featureId>.tgz.

Locally referenced Features Copy Section

Instead of publishing a Feature to an OCI registry, a Feature's source code may be referenced from a local folder. Locally referencing a Feature may be useful when first authoring a Feature.

A local Feature is referenced in the devcontainer's feature object relative to the folder containing the project's devcontainer.json.

Additional constraints exists when including local Features in a project:

• The project must have a .devcontainer/ folder at the root of the project workspace folder.

• A local Feature's source code must be contained within a sub-folder of the .devcontainer/ folder.

• The sub-folder name must match the Feature's id field.

• A local Feature may not be referenced by absolute path.

• The local Feature's sub-folder must contain at least a devcontainer-feature.json file and install.sh entrypoint script, mirroring the previously outlined file structure.

The relative path is provided using unix-style path syntax (e.g. ./myFeature) regardless of the host operating system.

An example project is illustrated below:

.
├── .devcontainer/
│   ├── localFeatureA/
│   │   ├── devcontainer-feature.json
│   │   ├── install.sh
│   │   └── ...
│   ├── localFeatureB/
│   │   ├── devcontainer-feature.json
│   │   ├── install.sh
│   │   └── ...
│   ├── devcontainer.json

devcontainer.json

{
        // ...
        "features": {
                "./localFeatureA": {},
                "./localFeatureB": {}
        }

}

File: docs/specs/devcontainer-id-variable.md Copy Chapter

Dev Container ID

This has now been implemented:

• Discussion issue: https://github.com/devcontainers/spec/issues/62
• CLI PR: https://github.com/devcontainers/cli/pull/242

Below is the original proposal.

Goal Copy Chapter

Allow features to refer to an identifier that is unique to the dev container they are installed into and that is stable across rebuilds.

E.g., the docker-in-docker feature needs a way to mount a volume per dev container.

Proposal Copy Chapter

The identifier will be referred to as ${devcontainerId} in the devcontainer.json and the feature metadata and that will be replaced with the dev container's id. It should only be used in parts of the configuration and metadata that is not used for building the image because that would otherwise prevent pre-building the image at a time when the dev container's id is not known yet. Excluding boolean, numbers and enum properties the properties supporting ${devcontainerId} in the devcontainer.json are: name, runArgs, initializeCommand, onCreateCommand, updateContentCommand, postCreateCommand, postStartCommand, postAttachCommand, workspaceFolder, workspaceMount, mounts, containerEnv, remoteEnv, containerUser, remoteUser, customizations. Excluding boolean, numbers and enum properties the properties supporting ${devcontainerId} in the feature metadata are: entrypoint, mounts, customizations.

Implementations can choose how to compute this identifier. They must ensure that it is unique among other dev containers on the same Docker host and that it is stable across rebuilds of dev containers. The identifier must only contain alphanumeric characters. We describe a way to do this below.

Label-based Implementation Copy Section

The following assumes that a dev container can be identified among other dev containers on the same Docker host by a set of labels on the container. Implementations may choose to follow this approach.

The identifier is derived from the set of container labels uniquely identifying the dev container. It is up to the implementation to choose these labels. For example, if the dev container is based on a local folder, the label could be named devcontainer.local_folder and have the local folder's path as its value.

For example, the ghcr.io/devcontainers/features/docker-in-docker feature could use the dev container id with:

{
    "id": "docker-in-docker",
    "version": "1.0.4",
    // ...
    "mounts": [
        {
            "source": "dind-var-lib-docker-${devcontainerId}",
            "target": "/var/lib/docker",
            "type": "volume"
        }
    ]
}

Label-based Computation Copy Section

• Input the labels as a JSON object with the object's keys being the label names and the object's values being the labels' values.
  • To ensure implementations get to the same result, the object keys must be sorted and any optional whitespace outside of the keys and values must be removed.
• Compute a SHA-256 hash from the UTF-8 encoded input string.
• Use a base-32 encoded representation left-padded with '0' to 52 characters as the result.

JavaScript implementation taking an object with the labels as argument and returning a string as the result:

const crypto = require('crypto');

function uniqueIdForLabels(idLabels) {
	const stringInput = JSON.stringify(idLabels, Object.keys(idLabels).sort()); // sort properties
	const bufferInput = Buffer.from(stringInput, 'utf-8');

	const hash = crypto.createHash('sha256')
		.update(bufferInput)
		.digest();

	const uniqueId = BigInt(`0x${hash.toString('hex')}`)
		.toString(32)
		.padStart(52, '0');
	return uniqueId;
}

File: docs/specs/devcontainer-lockfile.md Copy Chapter

Lockfiles

This has now been implemented:

• Discussion issue: https://github.com/devcontainers/spec/issues/236
• CLI PR: https://github.com/devcontainers/cli/issues/564

Below is the original proposal.

Goal Copy Chapter

Introduce a lockfile that records the exact version, download information and checksums for each feature listed in the devcontainer.json.

This will allow for:

• Improved reproducibility of image builds (installing "latest" of a tool will still have different outcomes as the tool publishes new releases).
• Improved cachability of image builds (image cache checksums will remain stable when the lockfile pins a feature to a particular version).
• Improved security by detecting when a feature's release artifact changes after its checksum was first recorded in the lockfile ("trust on first use").

Proposal Copy Chapter

(The following is inspired by NPM's package-lock.json and Yarn's yarn.lock.)

Each feature is recorded with the identifier and version it is referred to in the devcontainer.json as its key and the following properties as its values:

• resolved:
  • OCI feature: A qualified feature id with the sha256 hash (not the version number).
  • tarball feature: The https: URL used to download the feature.
• version: The full version number.
• integrity: sha256: followed by the hex encoded SHA-256 checksum of the download artifact.
• dependsOn: An array of feature identifiers equal to the feature's dependsOn keys in its devcontainer-feature.json including any version or checksum suffix. If the array is empty, this property can be omitted. For every feature listed here, the lockfile will also have a feature record.

The feature identifiers recorded as keys and in the dependsOn arrays are converted to lowercase to simplify lookups and comparisons. Note that the devcontainer.json and devcontainer-features.json may refer to the same features with different casing because these identifiers are not case-senstive.

Local features and the deprecated GitHub releases features are not recorded in the lockfile.

The lockfile is named devcontainer-lock.json, is located in the same folder as the devcontainer.json and contains a JSON object with a "features" property holding the above keys.

Example:

{
    "features": {
        "ghcr.io/devcontainers/features/node:1": {
            "version": "1.0.4",
            "resolved": "ghcr.io/devcontainers/features/node@sha256:567d704b3f4d3eca3acee51ded7c460a8395436d135d53d1175fb565daff42b8",
            "integrity": "sha256:567d704b3f4d3eca3acee51ded7c460a8395436d135d53d1175fb565daff42b8"
        },
        "https://mycustomdomain.com/devcontainer-feature-myfeature.tgz": {
            "version": "1.2.3",
            "resolved": "https://mycustomdomain.com/devcontainer-feature-myfeature.tgz",
            "integrity": "sha256:567d704b3f4d3eca3acee51ded7c460a8395436d135d53d1175fb565daff42b8"
        }
    }
}

Example with dependency:

{
  "features": {
    "ghcr.io/codspace/dependson/a": {
      "version": "1.2.1",
      "resolved": "ghcr.io/codspace/dependson/a@sha256:932027ef71da186210e6ceb3294c3459caaf6b548d2b547d5d26be3fc4b2264a",
      "integrity": "sha256:932027ef71da186210e6ceb3294c3459caaf6b548d2b547d5d26be3fc4b2264a",
      "dependsOn": [
        "ghcr.io/codspace/dependson/e:2"
      ]
    },
    "ghcr.io/codspace/dependson/e:2": {
      "version": "2.3.4",
      "resolved": "ghcr.io/codspace/dependson/e@sha256:9f36f159c70f8bebff57f341904b030733adb17ef12a5d58d4b3d89b2a6c7d5a",
      "integrity": "sha256:9f36f159c70f8bebff57f341904b030733adb17ef12a5d58d4b3d89b2a6c7d5a"
    }
  }
}

File: docs/specs/devcontainer-reference.md Copy Chapter

Dev Container Specification

The purpose of the Development Container Specification is to provide a way to enrich containers with the content and metadata necessary to enable development inside them. These container environments should be easy to use, create, and recreate.

A development container is a container in which a user can develop an application. Tools that want to implement this specification should provide a set of features/commands that give more flexibility to users and allow development containers to scale to large development groups.

An environment is defined as a logical instance of one or more development containers, along with any needed side-car containers. An environment is based on one set of metadata that can be managed as a single unit. Users can create multiple environments from the same configuration metadata for different purposes.

Metadata

Development containers allow one to define a repeatable development environment for a user or team of developers that includes the execution environment the application needs. A development container defines an environment in which you develop your application before you are ready to deploy. While deployment and development containers may resemble one another, you may not want to include tools in a deployment image that you use during development and you may need to use different secrets or other settings.

Furthermore, working inside a development container can require additional metadata to drive tooling or service experiences than you would normally need with a production container. Providing a structured and consistent form for this metadata is a core part of this specification.

devcontainer.json Copy Chapter

While the structure of this metadata is critical, it is also important to call out how this data can be represented on disk where appropriate. While other representations may be added over time, metadata can be stored in a JSON with Comments file called devcontainer.json today. Products using it should expect to find a devcontainer.json file in one or more of the following locations (in order of precedence):

• .devcontainer/devcontainer.json
• .devcontainer.json
• .devcontainer/<folder>/devcontainer.json (where <folder> is a sub-folder, one level deep)

It is valid that these files may exist in more than one location, so consider providing a mechanism for users to select one when appropriate.

Image Metadata Copy Chapter

Certain dev container metadata properties can be stored in an image label as an array of metadata snippets. This allows them to be stored in prebuilt images, such that, the image and its related configuration are self-contained. These contents should then be merged with any local devcontainer.json file contents at the time the container is created. An array is used so subsequent image builds can simply append changes to the array rather than attempting to merge at that point - which improves compatibility with arbitrary image build systems.

Metadata should be representative of with the following structure, using one entry per Dev Container Feature and devcontainer.json (see table below for the full list):

[
	{
		"id"?: string,
		"init"?: boolean,
		"privileged"?: boolean,
		"capAdd"?: string[],
		"securityOpt"?: string[],
		"entrypoint"?: string,
		"mounts"?: [],
		...
		"customizations"?: {
			...
		}
	},
	...
]

To simplify adding this metadata for other tools, we also support having a single top-level object with the same properties.

The metadata is added to the image as a devcontainer.metadata label with a JSON string value representing the above array or single object.

Merge Logic Copy Section

To apply the metadata together with a user's devcontainer.json at runtime the following merge logic by property is used. The table also notes which properties are currently supported coming from the devcontainer.json and which from the Feature metadata, this will change over time as we add more properties.

Variables in string values will be substituted at the time the value is applied. When the order matters, the devcontainer.json is considered last.

Notes Copy Section

• Passing the label as a LABEL instruction in the Dockerfile:
  • The size limit on Dockerfiles is around 1.3MB. The line length is limited to 65k characters.
  • Using one line per Feature should allow for making full use of these limits.
• Passing the label as a command line argument:
  • There is no size limit documented for labels, but the daemon returns an error when the request header is >500kb.
  • The 500kb limit is shared, so we cannot use a second label in the same build to avoid it.
  • If/when this becomes an issue we could embed the metadata as a file in the image (e.g. with a label indicating it).

Orchestration options

A core principle of this specification is to seek to enrich existing container orchestrator formats with development container metadata where appropriate rather than replacing them. As a result, the metadata schema includes a set of optional properties for interoperating with different orchestrators. Today, the specification includes scenario-specific properties for working without a container orchestrator (by directly referencing an image or Dockerfile) and for using Docker Compose as a simple multi-container orchestrator. At the same time, this specification leaves space for further development and implementation of other orchestrator mechanisms and file formats.

The following section describes the differences between those that are supported now.

Image based Copy Chapter

Image based configurations only reference an image that should be reachable and downloadable through docker pull commands. Logins and tokens required for these operations are execution environment specific. The only required parameter is image. The details are here.

Dockerfile based Copy Chapter

These configurations are defined as using a Dockerfile to define the starting point of the development containers. As with image based configurations, it is assumed that any base images are already reachable by Docker when performing a docker build command. The only required parameter in this case is the relative reference to the Dockerfile in build.dockerfile. The details are here.

There are multiple properties that allow users to control how docker build works:

• build.context
• build.args
• build.options
• build.target
• build.cacheFrom

Docker Compose based Copy Chapter

Docker Compose configurations use docker-compose (which may be Docker Compose V1 or aliased Docker Compose V2) to create and manage a set of containers required for an application. As with the other configurations, any images required for this operation are assumed to be reachable. The required parameters are:

• dockerComposeFile: the reference to the Docker Compose file(s) to be used.
• service: declares the main container that will be used for all other operations. Tools are assumed to also use this parameter to connect to the development container, although they can provide facilities to connect to the other containers as required by the user.
• runServices: an optional property that indicates the set of services in the docker-compose configuration that should be started or stopped with the environment.

It is important to note that image and dockerfile properties are not needed since Docker Compose supports them natively in the format.

Other options

In addition to the configuration options explained above, there are other settings that apply when creating development containers to facilitate their use by developers.

A complete list of available metadata properties and their purposes can be found in the devcontainer.json reference. However, we will describe the critical ones below in more detail.

Environment Variables Copy Chapter

Environment variables can be set at different points in the dev container lifecycle. With this in mind, development containers support two classes of environment variables:

• Container: These variables are part of the container when it is created and are available at all points in its lifecycle. This concept is native to containers and can be set in the container image itself, using containerEnv for image and Dockerfile scenarios or using orchestrator specific properties like env in Docker Compose files.
• Remote: These variables should be set by a development container supporting tool as part of configuring its runtime environment. Users can set these using the remoteEnv property and implementing tools or services may add their own for specific scenarios (e.g. secrets). These variables can change during the lifetime of the container, and are added after the container's ENTRYPOINT has fired.

The reason for this separation is it allows for the use of information not available at image build time and simplifies updating the environment for project/repository specific needs without modifying an image. With this in mind, it's important to note that implementing tools should also support the dynamic variable syntax described in the metadata reference document.

Another notable and important environment variable related property is userEnvProbe. Implementing tools should use this property to "probe" for expected environment variables using the specified type of shell. However, it does not specify that this type of shell needs to be used for all sub-processes (given the performance impact). Instead, "probed" environment variables should be merged with Remote environment variables for any processes the implementer injects after the container is created. This allows implementors to emulate developer expected behaviors around values added to their profile and rc files.

Mounts Copy Chapter

Mounts allow containers to have access to the underlying machine, share data between containers and to persist information between development containers.

A default mount should be included so that the source code is accessible from inside the container. Source code is stored outside of the container so that a developer's in-flight edits can be extracted, or a new container created in the event a container no longer starts.

While this "workspace mount" is often a "bind" mount, this is not a requirement of this specification. It is also important to note that these mounts may be "bind" mounts that connect to the underlying filesystem and thus cloud environments might not have access to the same data as a local machine.

Inside the container this mount defaults to /workspace.

workspaceFolder and workspaceMount Copy Chapter

The default mount point for the source code can be set with the workspaceMount property for image and dockerfile scenarios or using the built in mounts property in Docker Compose files. This folder should point to the root of a repository (where the .git folder is found) so that source control operations work correctly inside the container.

The workspaceFolder can then be set to the default folder inside the container that should used in the container. Typically this is either the mount point in the container, or a sub-folder under it. Allowing a sub-folder to be used is particularly important for monorepos given you need the .git folder to interact with source control but developers are typically interacting with a specific sub-project within the overall repository.

See workspaceMount and workspaceFolder for reference.

Users Copy Chapter

Users control the permissions of applications executed in the containers, allowing the developer to control them. The specification takes into account two types of user definitions:

• Container User: The user that will be used for all operations that run inside a container. This concept is native to containers. It may be set in the container image, using the containerUser property for image and dockerfile scenarios, or using an orchestratric specific property like user property in Docker Compose files.
• Remote User: Used to run the lifecycle scripts inside the container. This is also the user tools and editors that connect to the container should use to run their processes. This concept is not native to containers. Set using the remoteUser property in all cases and defaults to the container user.

This separation allows the ENTRYPOINT for the image to execute with different permissions than the developer and allows for developers to switch users without recreating their containers.

Lifecycle

A development environment goes through different lifecycle events during its use in the outer and inner loop of development.

• Configuration Validation
• Environment Creation
• Environment Stop
• Environment Resume

Configuration Validation Copy Chapter

The exact steps required to validate configuration can vary based on exactly where the development container metadata is persisted. However, when considering a devcontainer.json file, the following validation should occur:

1. Validate that a workspace source folder has been provided. It is up to the implementing tool to determine what to do if no source folder is provided.
2. Search for a devcontainer.json file in one of the locations above in the workspace source folder.
3. If no devcontainer.json is found, it is up to the implementing tool or service to determine what to do. This specification does not dictate this behavior.
4. Validate that the metadata (for example devcontainer.json) contains all parameters required for the selected configuration type.

Environment Creation Copy Chapter

The creation process goes through the steps necesarry to go from the user configuration to a working environment that is ready to be used.

Initialization Copy Section

During this step, the following is executed:

• Validate access to the container orchestrator specified by the configuration.
• Execution of initializeCommand.

Image creation Copy Section

The first part of environment creation is generating the final image(s) that the development containers are going to use. This step is orchestrator dependent and can consist of just pulling a Docker image, running Docker build, or docker-compose build. Additionally, this step is useful on its own since it permits the creation of intermediate images that can be uploaded and used by other users, thus cutting down on creation time. It is encouraged that tools implementing this specification give access to a command that just executes this step.

This step executes the following tasks:

1. Configuration Validation
2. Pull/build/execute of the defined container orchestration format to create images.
3. Validate the result of these operations.

Container Creation Copy Section

After image creation, containers are created based on that image and setup.

This step executes the following tasks:

1. [Optional] Perform any required user UID/GID sync'ing (more next)
2. Create the container(s) based on the properties specified above.
3. Validate the container(s) were created successfully.

Note that container mounts, environment variables, and user configuration should be applied at this point. However, remote user and environment variable configuration should not be.

UID/GID sync'ing is an optional task for Linux (only) and that executes if the updateRemoteUserUID property is set to true and a containerUser or remoteUser is specified. In this case, an image update should be made prior to creating the container to set the specified user's UID and GID to match the current local user’s UID/GID to avoid permission problems with bind mounts. Implementations may skip this task if they do not use bind mounts on Linux, or use a container engine that does this translation automatically.

Post Container Creation Copy Section

At the end of the container creation step, a set of commands are executed inside the main container:

• onCreateCommand, updateContentCommand and postCreateCommand. This set of commands is executed in sequence on a container the first time it's created and depending on the creation parameters received. You can learn more in the documentation on lifecycle scripts. By default, postCreateCommand is executed in the background after reporting the successful creation of the development environment.
• If the waitFor property is defined, then execution should block until all commands in the sequence up to the specified property have executed. This property defaults to updateContentCommand.

Remote environment variables and user configuration should be applied to all created processes in the container (inclusive of userEnvProbe).

Implementation specific steps Copy Section

After these steps have been executed, any implementation specific commands can safely execute. Specifically, any processes required by the implementation to support other properties in this specification should be started at this point. These may occur in parallel to any non-blocking, background post-container creation commands (as dictated by the waitFor property).

Any user facing processes should have remote environment variables and user configuration applied (inclusive of userEnvProbe).

For example, in the CLI reference implementation, this is the point in which anything executed with devcontainer exec would run.

Typically, this is also the step where implementors would apply config or settings from the customizations section of the dev container metadata (e.g. VS Code installs extensions based on the customizations.vscode.extensions property). Examples of these can be found in the supporting tools section reference. However, applying these at this point is not strictly required or mandated by this specification.

Once these final steps have occurred, implementing tools or services may connect to the environment as they see fit.

Environment Stop Copy Chapter

The intention of this step is to ensure all containers are stopped correctly based on the appropriate orchestrator specific steps to ensure no data is lost. It is up to the implementing tool or service to determine when this event should happen.

Environment Resume Copy Chapter

While it is not a strict requirement to keep a development container after it has been stopped, this is the most common scenario.

To resume the environment from a stopped state:

1. Restart all related containers.
2. Follow the appropriate implementation specific steps.
3. Additionally, execute the postStartCommand and postAttachCommand in the container.

Like during the create process, remote environment variables and user configuration should be applied to all created processes in the container (inclusive of userEnvProbe).

Parallel lifecycle script execution Copy Chapter

Dev containers support a single command for each of its lifecycle scripts. While serial execution of multiple commands can be achieved with ;, &&, etc., parallel execution deserves first-class support.

All lifecycle scripts have been extended to support object types. The key of the object will be a unique name for the command and the value will be the string or array command. Each command must exit successfully for the stage to be considered successful.

Each entry in the object will be run in parallel during that lifecycle step.

Example Copy Section

{
  "postCreateCommand": {
    "server": "npm start",
    "db": ["mysql", "-u", "root", "-p", "my database"]
  }
}

Definitions

**Project Workspace Folder**

The project workspace folder is where an implementing tool should begin to search for devcontainer.json files. If the target project on disk is using git, the project workspace folder is typically the root of the git repository.

File: docs/specs/devcontainer-templates.md Copy Chapter

Dev Container Templates reference

Development Container Templates are source files packaged together that encode configuration for a complete development environment. A Template can be used in a new or existing project, and a supporting tool will use the configuration from the Template to build a development container.

The configuration is placed in .devcontainer/devcontainer.json which can also reference other files within the Template. A Template can also provide additional source files (e.g. boilerplate code or a lifecycle script.

Template metadata is captured by a devcontainer-template.json file in the root folder of the Template.

Folder structure Copy Chapter

A single Template is a folder with at least a devcontainer-template.json and .devcontainer/devcontainer.json. Additional files are permitted and are packaged along side the required files.

+-- template
|   +-- devcontainer-template.json
|   +-- .devcontainer
|       +-- devcontainer.json
|       +-- (other files)
|   +-- (other files)

devcontainer-template.json properties Copy Chapter

The devcontainer-template.json file defines information about the Template to be used by any supporting tools.

The properties of the file are as follows:

The `options` property Copy Section

The options property contains a map of option IDs and their related configuration settings. These options are used by the supporting tools to prompt the user to choose from different Template configuration options. The tools would replace the option ID with the selected value in all the files (within the sub-directory of the Template). This replacement would happen before dropping the .devcontainer/devcontainer.json and other files (within the sub-directory of the Template) required to containerize your project. See option resolution for more details. For example:

{
  "options": {
    "optionId": {
      "type": "string",
      "description": "Description of the option",
      "proposals": ["value1", "value2"],
      "default": "value1"
    }
  }
}

Note: The options must be unique for every devcontainer-template.json

The `optionalPaths` property Copy Section

Before applying a Template, tooling must inspect the optionalPaths property of a Template and prompt the user on whether each file or folder should be included in the resulting output workspace folder. A path is relative to the root of the Template source directory.

• For a single file, provide the full relative path (without any leading or trailing path delimiters).
• For a directory, provide the full relative path with a trailing slash and asterisk (/*) appended to the path. The directory and its children will be recursively ignored.

Examples are shown below:

{
    "id": "cpp",
    "version": "3.0.0",
    "name": "C++",
    "description": "Develop C++ applications",
    "optionalPaths": [
         "GETTING-STARTED.md",                 // Single file
         "example-project-1/MyProject.csproj", // Single file in nested directory
         ".github/*"                           // Entire recursive contents of directory
     ]
}

Referencing a Template Copy Section

The id format (<oci-registry>/<namespace>/<template>[:<semantic-version>]) dictates how a supporting tool will locate and download a given Template from an OCI registry. For example:

• ghcr.io/user/repo/go
• ghcr.io/user/repo/go:1
• ghcr.io/user/repo/go:latest

The registry must implement the OCI Artifact Distribution Specification. Some implementors can be found here.

Versioning Copy Chapter

Each Template is individually versioned according to the semver specification. The version property in the respective devcontainer-template.json file is updated to increment the Template's version.

Tooling that handles releasing Templates will not republish Templates if that exact version has already been published; however, tooling must republish major and minor versions in accordance with the semver specification.

Release Copy Chapter

For information on distributing Templates, see devcontainer-templates-distribution.md.

Option Resolution Copy Section

A Template's options property is used by a supporting tool to prompt for different configuration options. A supporting tool will parse the options object provided by the user. If a value is selected for a Template, it will be replaced in the files (within the sub-directory of the Template).

Option resolution example Copy Section

Consider a java Template with the following folder structure:

+-- java
|   +-- devcontainer-template.json
|   +-- .devcontainer
|       +-- devcontainer.json

Suppose the java Template has the following options parameters declared in the devcontainer-template.json file:

// ...
"options": {
    "imageVariant": {
        "type": "string",
        "description": "Specify version of java.",
        "proposals": [
          "17-bullseye",
          "17-buster",
          "11-bullseye",
          "11-buster",
          "17",
          "11"
        ],
			  "default": "17-bullseye"
    },
    "nodeVersion": {
        "type": "string", 
        "proposals": [
          "latest",
          "16",
          "14",
          "10",
          "none"
        ],
        "default": "latest",
        "description": "Specify version of node, or 'none' to skip node installation."
    },
    "installMaven": {
        "type": "boolean", 
        "description": "Install Maven, a management tool for Java.",
        "default": "false"
    },
}

...and it has the following .devcontainer/devcontainer.json file:

{
	"name": "Java",
	"image": "mcr.microsoft.com/devcontainers/java:0-${templateOption:imageVariant}",
	"features": {
		"ghcr.io/devcontainers/features/node:1": {
			"version": "${templateOption:nodeVersion}",
      "installMaven": "${templateOption:installMaven}"
		}
	},
//	...
}

A user tries to add the java Template to their project using the supporting tools and selects 17-bullseye when prompted for "Specify version of Go" and the default values when prompted for "Specify version of node, or 'none' to skip node installation" and "Install Maven, a management tool for Java".

The supporting tool could then use a string replacer for all the files within the sub-directory of the Template. In this example, .devcontainer/devcontainer.json needs to be modified and hence, the inputs can provided to it as follows:

{
  imageVariant:"17-bullseye",
  nodeVersion: "latest",
  installMaven: "false"
}

The modified .devcontainer/devcontainer.json will be as follows:

{
	"name": "Go",
	"image": "mcr.microsoft.com/devcontainers/go:0-17-bullseye",
	"features": {
		"ghcr.io/devcontainers/features/node:1": {
			"version": "latest",
      "installMaven": "false"
		}
	},
	...
}

The modified .devcontainer/devcontainer.json would be dropped into any existing folder as a starting point for containerizing your project.

File: docs/specs/devcontainer-templates-distribution.md Copy Chapter

Distribution and Discovery of Dev Container Templates

TL;DR Check out the quick start repository to get started on distributing your own Dev Container Templates.

This specification defines a pattern where community members and organizations can author and self-publish Dev Container Templates.

Goals include:

• For Template authors, create a "self-service" way to publish a Template, either publicly or privately, that is not centrally controlled.
• Provide the ability to standardize publishing such that supporting tools may implement their own mechanism to aid Template discoverability as they see fit.

Source code Copy Chapter

A Template's source code is stored in a git repository.

For ease of authorship and maintenance, [1..n] Templates can share a single git repository. This set of Templates is referred to as a "collection," and will share the same devcontainer-collection.json file and "namespace" (eg. <owner>/<repo>).

Note: Templates and Features should be placed in different git repositories.

Source code for a set of Templates follows the example file structure below:

.
├── README.md
├── src
│   ├── dotnet
│   │   ├── devcontainer-template.json
│   │   ├── .devcontainer
│   │       ├── devcontainer.json
│   │       └── ...
│   │   ├── ...
|   ├
│   ├── docker-from-docker
│   │   ├── devcontainer-template.json
│   │   ├── .devcontainer
│   │       ├── devcontainer.json
│   │       ├── Dockerfile
│   │       └── ...
│   │   ├── ...
|   ├
│   ├── go-postgres
│   │   ├── devcontainer-template.json
│   │   ├── .devcontainer
│   │       ├── devcontainer.json
│   │       ├── docker-compose.yml
│   │       ├── Dockerfile
│   │       └── ...
│   │   ├── ...

...where src is a directory containing a sub-folder with the name of the Template (e.g. src/dotnet or src/docker-from-docker) with at least a file named devcontainer-template.json that contains the Template metadata, and a .devcontainer/devcontainer.json that the supporting tools will drop into an existing project or folder.

Each sub-directory should be named such that it matches the id field of the devcontainer-template.json. Other files can also be included in the Templates's sub-directory, and will be included during the packaging step alongside the two required files. Any files that are not part of the Templates's sub-directory (e.g. outside of src/dotnet) will not be included in the packaging step.

Versioning Copy Chapter

Each Template is individually versioned according to the semver specification. The version property in the respective devcontainer-template.json file is parsed to determine if the Template should be republished.

Tooling that handles publishing Templates will not republish Templates if that exact version has already been published; however, tooling must republish major and minor versions in accordance with the semver specification.

Packaging Copy Chapter

Templates are distributed as tarballs. The tarball contains the entire contents of the Template sub-directory, including the devcontainer-template.json, .devcontainer/devcontainer.json, and any other files in the directory.

The tarball is named devcontainer-template-.tgz, where `` is the Templates's id field.

A reference implementation for packaging and distributing Templates is provided as a GitHub Action (https://github.com/devcontainers/action).

devcontainer-collection.json Copy Section

The devcontainer-collection.json is an auto-generated metadata file.

Each Template's devcontainer-template.json metadata file is appended into the templates top-level array.

Distribution Copy Chapter

There are several supported ways to distribute Templates. Distribution is handled by the implementing packaging tool such as the Dev Container CLI or Dev Container Publish GitHub Action.

A user can add a Template in to their projects as defined by the supporting Tools.

OCI Registry Copy Section

An OCI registry that implements the OCI Artifact Distribution Specification serves as the primary distribution mechanism for Templates.

Each packaged Template is pushed to the registry following the naming convention <registry>/<namespace>/[:version], where version is the major, minor, and patch version of the Template, according to the semver specification.

Note: namespace is a unique identifier for the collection of Templates and must be different than the collection of Features. There are no strict rules for namespace; however, one pattern is to set namespace equal to source repository's <owner>/<repo>.

A custom media type application/vnd.devcontainers and application/vnd.devcontainers.layer.v1+tar are used as demonstrated below.

For example, the go Template in the devcontainers/templates namespace at version 1.2.3 would be pushed to the ghcr.io OCI registry.

Note: The example below uses oras for demonstration purposes. A supporting tool should directly implement the required functionality from the aforementioned OCI artifact distribution specification.

# ghcr.io/devcontainers/templates/go:1
REGISTRY=ghcr.io
NAMESPACE=devcontainers/templates
TEMPLATE=go

ARTIFACT_PATH=devcontainer-template-go.tgz

for VERSION in 1  1.2  1.2.3  latest
do
        oras push ${REGISTRY}/${NAMESPACE}/${TEMPLATE}:${VERSION} \
                --config /dev/null:application/vnd.devcontainers \
                        ./${ARTIFACT_PATH}:application/vnd.devcontainers.layer.v1+tar
done

The "namespace" is the globally identifiable name for the collection of Templates (e.g. owner/repo for the source code's git repository).

The auto-generated devcontainer-collection.json is pushed to the registry with the same namespace as above and no accompanying template name. The collection file is always tagged as latest.

# ghcr.io/devcontainers/templates
REGISTRY=ghcr.io
NAMESPACE=devcontainers/templates

oras push ${REGISTRY}/${NAMESPACE}:latest \
        --config /dev/null:application/vnd.devcontainers \
                            ./devcontainer-collection.json:application/vnd.devcontainers.collection.layer.v1+json

2. Official Technical Reference & Guides (devcontainers/devcontainers.github.io) Copy Chapter

Development Containers Website

This repo holds the website for the Development Containers Specification.

You may view the site at containers.dev.

Building Copy Chapter

If you'd like to build and preview the site yourself, we make it as smooth as possible through a dev container in this repo!

Dev container Copy Section

You may build GitHub Pages sites with Jekyll, which is a Ruby gem. You could manually install these tools on your machine, or you can easily get started with the setup you already need through a dev container!

You may review this repo's dev container in the .devcontainer folder.

It is from this Jekyll Dev Container Template.

Steps to build and run Copy Section

• Clone or open this repo in the dev container-supporting editor of your choosing.
  • You may review supporting tools and services here.
• Reopen this repo in the dev container, so that the container builds and you may develop inside it using the included tools.
• Once the dev container finishes building, execute the following command in your dev container to start the site: bundle exec jekyll serve
• Check out the site! http://localhost:4000/containers.dev/

Feedback and contributing Copy Chapter

If you'd like to provide feedback on or contribute to the dev containers website, please feel free to open an issue or PR in this repo.

If you'd like to add your own Features or Templates collection to the community index, please read the Authoring a Dev Container Feature guide first — in particular the Adding Features to the Index section, which explains how to correctly format your collection-index.yml entry.

For issues on and contributions to the dev container specification itself, please visit the Dev Containers Spec repo.

License Copy Chapter

License for this repository: https://github.com/devcontainers/containers.dev/blob/gh-pages/LICENSE.