How Build Inputs Work

A build input provides source content for builds to operate on. There are several ways to provide source in OpenShift Container Platform. In order of precedence:

Different inputs can be combined into a single build. As the inline Dockerfile takes precedence, it can overwrite any other file named Dockerfile provided by another input. Binary input and Git repositories are mutually exclusive inputs.

Input secrets are useful for when you do not want certain resources or credentials used during a build to be available in the final application image produced by the build, or want to consume a value that is defined in a Secret resource. External artifacts can be used to pull in additional files that are not available as one of the other build input types.

Whenever a build is run:

  1. A working directory is constructed and all input content is placed in the working directory. For example, the input Git repository is cloned into the working directory, and files specified from input images are copied into the working directory using the target path.

  2. The build process changes directories into the contextDir, if one is defined.

  3. The inline Dockerfile, if any, is written to the current directory.

  4. The content from the current directory is provided to the build process for reference by the Dockerfile, custom builder logic, or assemble script. This means any input content that resides outside the contextDir will be ignored by the build.

The following example of a source definition includes multiple input types and an explanation of how they are combined. For more details on how each input type is defined, see the specific sections for each input type.

source:
  git:
    uri: https://github.com/openshift/ruby-hello-world.git (1)
  images:
  - from:
      kind: ImageStreamTag
      name: myinputimage:latest
      namespace: mynamespace
    paths:
    - destinationDir: app/dir/injected/dir (2)
      sourcePath: /usr/lib/somefile.jar
  contextDir: "app/dir" (3)
  dockerfile: "FROM centos:7\nRUN yum install -y httpd" (4)
1 The repository to be cloned into the working directory for the build.
2 /usr/lib/somefile.jar from myinputimage will be stored in <workingdir>/app/dir/injected/dir.
3 The working directory for the build will become <original_workingdir>/app/dir.
4 A Dockerfile with this content will be created in <original_workingdir>/app/dir, overwriting any existing file with that name.

Dockerfile Source

When a dockerfile value is supplied, the content of this field will be written to disk as a file named Dockerfile. This is done after other input sources are processed, so if the input source repository contains a Dockerfile in the root directory, it will be overwritten with this content.

The typical use for this field is to provide a Dockerfile to a Docker strategy build.

The source definition is part of the spec section in the BuildConfig:

source:
  dockerfile: "FROM centos:7\nRUN yum install -y httpd" (1)
1 The dockerfile field contains an inline Dockerfile that will be built.

Image Source

Additional files can be provided to the build process via images. Input images are referenced in the same way the From and To image targets are defined. This means both container images and image stream tags can be referenced. In conjunction with the image, you must provide one or more path pairs to indicate the path of the files or directories to copy the image and the destination to place them in the build context.

The source path can be any absolute path within the image specified. The destination must be a relative directory path. At build time, the image will be loaded and the indicated files and directories will be copied into the context directory of the build process. This is the same directory into which the source repository content (if any) is cloned. If the source path ends in /. then the content of the directory will be copied, but the directory itself will not be created at the destination.

Image inputs are specified in the source definition of the BuildConfig:

source:
  git:
    uri: https://github.com/openshift/ruby-hello-world.git
  images: (1)
  - from: (2)
      kind: ImageStreamTag
      name: myinputimage:latest
      namespace: mynamespace
    paths: (3)
    - destinationDir: injected/dir (4)
      sourcePath: /usr/lib/somefile.jar (5)
  - from:
      kind: ImageStreamTag
      name: myotherinputimage:latest
      namespace: myothernamespace
    pullSecret: mysecret (6)
    paths:
    - destinationDir: injected/dir
      sourcePath: /usr/lib/somefile.jar
1 An array of one or more input images and files.
2 A reference to the image containing the files to be copied.
3 An array of source/destination paths.
4 The directory relative to the build root where the build process can access the file.
5 The location of the file to be copied out of the referenced image.
6 An optional secret provided if credentials are needed to access the input image.

This feature is not supported for builds using the Custom Strategy.

Git Source

When specified, source code will be fetched from the location supplied.

If an inline Dockerfile is supplied, it will overwrite the Dockerfile (if any) in the contextDir of the Git repository.

The source definition is part of the spec section in the BuildConfig:

source:
  git: (1)
    uri: "https://github.com/openshift/ruby-hello-world"
    ref: "master"
  contextDir: "app/dir" (2)
  dockerfile: "FROM openshift/ruby-22-centos7\nUSER example" (3)
1 The git field contains the URI to the remote Git repository of the source code. Optionally, specify the ref field to check out a specific Git reference. A valid ref can be a SHA1 tag or a branch name.
2 The contextDir field allows you to override the default location inside the source code repository where the build looks for the application source code. If your application exists inside a sub-directory, you can override the default location (the root folder) using this field.
3 If the optional dockerfile field is provided, it should be a string containing a Dockerfile that overwrites any Dockerfile that may exist in the source repository.

If the ref field denotes a pull request, the system will use a git fetch operation and then checkout FETCH_HEAD.

When no ref value is provided, OpenShift Container Platform performs a shallow clone (--depth=1). In this case, only the files associated with the most recent commit on the default branch (typically master) are downloaded. This results in repositories downloading faster, but without the full commit history. To perform a full git clone of the default branch of a specified repository, set ref to the name of the default branch (for example master).

Using a Proxy

If your Git repository can only be accessed using a proxy, you can define the proxy to use in the source section of the BuildConfig. You can configure both a HTTP and HTTPS proxy to use. Both fields are optional. Domains for which no proxying should be performed can also be specified via the NoProxy field.

Your source URI must use the HTTP or HTTPS protocol for this to work.

source:
  git:
    uri: "https://github.com/openshift/ruby-hello-world"
    httpProxy: http://proxy.example.com
    httpsProxy: https://proxy.example.com
    noProxy: somedomain.com, otherdomain.com

For Pipeline strategy builds, given the current restrictions with the Git plug-in for Jenkins, any Git operations through the Git plug-in will not leverage the HTTP or HTTPS proxy defined in the BuildConfig. The Git plug-in only will use the the proxy configured in the Jenkins UI at the Plugin Manager panel. This proxy will then be used for all git interactions within Jenkins, across all jobs. You can find instructions on how to configure proxies through the Jenkins UI at JenkinsBehindProxy.

Source Clone Secrets

Builder pods require access to any Git repositories defined as source for a build. Source clone secrets are used to provide the builder pod with access it would not normally have access to, such as private repositories or repositories with self-signed or untrusted SSL certificates.

The following source clone secret configurations are supported.

You can also use combinations of these configurations to meet your specific needs.

Builds are run with the builder service account, which must have access to any source clone secrets used. Access is granted with the following command:

$ oc secrets link builder mysecret

Limiting secrets to only the service accounts that reference them is disabled by default. This means that if serviceAccountConfig.limitSecretReferences is set to false (the default setting) in the master configuration file, linking secrets to a service is not required.

Automatically Adding a Source Clone Secret to a Build Configuration

When a BuildConfig is created, OpenShift Container Platform can automatically populate its source clone secret reference. This behaviour allows the resulting Builds to automatically use the credentials stored in the referenced Secret to authenticate to a remote Git repository, without requiring further configuration.

To use this functionality, a Secret containing the Git repository credentials must exist in the namespace in which the BuildConfig will later be created. This Secret must additionally include one or more annotations prefixed with build.openshift.io/source-secret-match-uri-. The value of each of these annotations is a URI pattern, defined as follows. When a BuildConfig is created without a source clone secret reference and its Git source URI matches a URI pattern in a Secret annotation, OpenShift Container Platform will automatically insert a reference to that Secret in the BuildConfig.

A URI pattern must consist of:

  • a valid scheme (*://, git://, http://, https:// or ssh://).

  • a host (* or a valid hostname or IP address optionally preceded by *.).

  • a path (/* or / followed by any characters optionally including * characters).

In all of the above, a * character is interpreted as a wildcard.

URI patterns only match Git source URIs which are conformant to RFC3986. For example, https://github.com/openshift/origin.git. They do not match the alternate SSH style that Git also uses. For example, git@github.com:openshift/origin.git.

It is not valid to attempt to express a URI pattern in the alternate style, or to include a username/password component in a URI pattern.

If multiple Secrets match the Git URI of a particular BuildConfig, OpenShift Container Platform will select the secret with the longest match. This allows for basic overriding, as in the following example.

The following fragment shows two partial source clone secrets, the first matching any server in the domain mycorp.com accessed by HTTPS, and the second overriding access to servers mydev1.mycorp.com and mydev2.mycorp.com:

kind: Secret
apiVersion: v1
metadata:
  name: matches-all-corporate-servers-https-only
  annotations:
    build.openshift.io/source-secret-match-uri-1: https://*.mycorp.com/*
data:
  ...

kind: Secret
apiVersion: v1
metadata:
  name: override-for-my-dev-servers-https-only
  annotations:
    build.openshift.io/source-secret-match-uri-1: https://mydev1.mycorp.com/*
    build.openshift.io/source-secret-match-uri-2: https://mydev2.mycorp.com/*
data:
  ...

Add a build.openshift.io/source-secret-match-uri- annotation to a pre-existing secret using:

$ oc annotate secret mysecret \
    'build.openshift.io/source-secret-match-uri-1=https://*.mycorp.com/*'

Manually Adding Source Clone Secrets

Source clone secrets can be added manually to a build configuration by adding a sourceSecret field to the source section inside the BuildConfig and setting it to the name of the secret that you created (basicsecret, in this example).

apiVersion: "v1"
kind: "BuildConfig"
metadata:
  name: "sample-build"
spec:
  output:
    to:
      kind: "ImageStreamTag"
      name: "sample-image:latest"
  source:
    git:
      uri: "https://github.com/user/app.git"
    sourceSecret:
      name: "basicsecret"
  strategy:
    sourceStrategy:
      from:
        kind: "ImageStreamTag"
        name: "python-33-centos7:latest"

You can also use the oc set build-secret command to set the source clone secret on an existing build configuration:

$ oc set build-secret --source bc/sample-build basicsecret

Defining Secrets in the BuildConfig provides more information on this topic.

.Gitconfig File

If the cloning of your application is dependent on a .gitconfig file, then you can create a secret that contains it, and then add it to the builder service account, and then your BuildConfig.

To create a secret from a .gitconfig file:

$ oc create secret generic <secret_name> --from-file=<path/to/.gitconfig>

SSL verification can be turned off if sslVerify=false is set for the http section in your .gitconfig file:

[http]
        sslVerify=false

Basic Authentication

Basic authentication requires either a combination of --username and --password, or a token to authenticate against the SCM server.

Create the secret first before using the user name and password to access the private repository:

$ oc create secret generic <secret_name> \
    --from-literal=username=<user_name> \
    --from-literal=password=<password> \
    --type=kubernetes.io/basic-auth

To create a basic authentication secret with a token:

$ oc create secret generic <secret_name> \
    --from-literal=password=<token> \
    --type=kubernetes.io/basic-auth

SSH Key Authentication

SSH key based authentication requires a private SSH key.

The repository keys are usually located in the $HOME/.ssh/ directory, and are named id_dsa.pub, id_ecdsa.pub, id_ed25519.pub, or id_rsa.pub by default. Generate SSH key credentials with the following command:

$ ssh-keygen -t rsa -C "your_email@example.com"

Creating a passphrase for the SSH key prevents OpenShift Container Platform from building. When prompted for a passphrase, leave it blank.

Two files are created: the public key and a corresponding private key (one of id_dsa, id_ecdsa, id_ed25519, or id_rsa). With both of these in place, consult your source control management (SCM) system’s manual on how to upload the public key. The private key is used to access your private repository.

Before using the SSH key to access the private repository, create the secret first:

$ oc create secret generic <secret_name> \
    --from-file=ssh-privatekey=<path/to/ssh/private/key> \
    --type=kubernetes.io/ssh-auth

Trusted Certificate Authorities

The set of TLS certificate authorities that are trusted during a git clone operation are built into the OpenShift Container Platform infrastructure images. If your Git server uses a self-signed certificate or one signed by an authority not trusted by the image, you have several options.

  1. Create a secret with a CA certificate file (recommended).

    A secret containing a CA certificate in a key named ca.crt will automatically be used by Git to trust your self-signed or otherwise un-trusted TLS certificate during the git clone operation. Using this method is significantly more secure than disabling Git’s SSL verification, which accepts any TLS certificate that is presented.

    $ oc create secret generic mycert --from-file=ca.crt=</path/to/file> (1)
    1 The key name ca.crt must be used.
  2. Disable Git TLS verification.

    You can disable Git’s TLS verification by setting the GIT_SSL_NO_VERIFY environment variable to true in the appropriate strategy section of your build configuration. You can use the oc set env command to manage BuildConfig environment variables.

Combinations

Below are several examples of how you can combine the above methods for creating source clone secrets for your specific needs.

  1. To create an SSH-based authentication secret with a .gitconfig file:

    $ oc create secret generic <secret_name> \
        --from-file=ssh-privatekey=<path/to/ssh/private/key> \
        --from-file=<path/to/.gitconfig> \
        --type=kubernetes.io/ssh-auth
  2. To create a secret that combines a .gitconfig file and CA certificate:

    $ oc create secret generic <secret_name> \
        --from-file=ca.crt=<path/to/certificate> \
        --from-file=<path/to/.gitconfig>
  3. To create a basic authentication secret with a CA certificate file:

    $ oc create secret generic <secret_name> \
        --from-literal=username=<user_name> \
        --from-literal=password=<password> \
        --from-file=ca-cert=</path/to/file> \
        --type=kubernetes.io/basic-auth
  4. To create a basic authentication secret with a .gitconfig file:

    $ oc create secret generic <secret_name> \
        --from-literal=username=<user_name> \
        --from-literal=password=<password> \
        --from-file=</path/to/.gitconfig> \
        --type=kubernetes.io/basic-auth
  5. To create a basic authentication secret with a .gitconfig file and CA certificate file:

    $ oc create secret generic <secret_name> \
        --from-literal=username=<user_name> \
        --from-literal=password=<password> \
        --from-file=</path/to/.gitconfig> \
        --from-file=ca-cert=</path/to/file> \
        --type=kubernetes.io/basic-auth

Binary Source

Streaming content in binary format from a local file system to the builder is called a Binary type build. The corresponding value of BuildConfig.spec.source.type is Binary for such builds.

This source type is unique in that it is leveraged solely based on your use of the oc start-build.

Binary type builds require content to be streamed from the local file system, so automatically triggering a binary type build (e.g. via an image change trigger) is not possible, because the binary files cannot be provided. Similarly, you cannot launch binary type builds from the web console.

To utilize binary builds, invoke oc start-build with one of these options:

  • --from-file: The contents of the file you specify are sent as a binary stream to the builder. You can also specify a URL to a file. Then, the builder stores the data in a file with the same name at the top of the build context.

  • --from-dir and --from-repo: The contents are archived and sent as a binary stream to the builder. Then, the builder extracts the contents of the archive within the build context directory. With --from-dir, you can also specify a URL to an archive, which will be extracted.

  • --from-archive: The archive you specify is sent to the builder, where it is extracted within the build context directory. This option behaves the same as --from-dir; an archive is created on your host first, whenever the argument to these options is a directory.

In each of the above cases:

  • If your BuildConfig already has a Binary source type defined, it will effectively be ignored and replaced by what the client sends.

  • If your BuildConfig has a Git source type defined, it is dynamically disabled, since Binary and Git are mutually exclusive, and the data in the binary stream provided to the builder takes precedence.

Instead of a file name, you can pass a URL with HTTP or HTTPS schema to --from-file and --from-archive. When using --from-file with a URL, the name of the file in the builder image is determined by the Content-Disposition header sent by the web server, or the last component of the URL path if the header is not present. No form of authentication is supported and it is not possible to use custom TLS certificate or disable certificate validation.

When using oc new-build --binary=true, the command ensures that the restrictions associated with binary builds are enforced. The resulting BuildConfig will have a source type of Binary, meaning that the only valid way to run a build for this BuildConfig is to use oc start-build with one of the --from options to provide the requisite binary data.

The dockerfile and contextDir source options have special meaning with binary builds.

dockerfile can be used with any binary build source. If dockerfile is used and the binary stream is an archive, its contents serve as a replacement Dockerfile to any Dockerfile in the archive. If dockerfile is used with the --from-file argument, and the file argument is named dockerfile, the value from dockerfile replaces the value from the binary stream.

In the case of the binary stream encapsulating extracted archive content, the value of the contextDir field is interpreted as a subdirectory within the archive, and, if valid, the builder changes into that subdirectory before executing the build.

Input Secrets

In some scenarios, build operations require credentials to access dependent resources, but it is undesirable for those credentials to be available in the final application image produced by the build. You can define input secrets for this purpose.

For example, when building a Node.js application, you can set up your private mirror for Node.js modules. In order to download modules from that private mirror, you have to supply a custom .npmrc file for the build that contains a URL, user name, and password. For security reasons, you do not want to expose your credentials in the application image.

This example describes Node.js, but you can use the same approach for adding SSL certificates into the /etc/ssl/certs directory, API keys or tokens, license files, and more.

Adding Input Secrets

To add an input secret to an existing BuildConfig:

  1. Create the secret, if it does not exist:

    $ oc create secret generic secret-npmrc \
        --from-file=.npmrc=<path/to/.npmrc>

    This creates a new secret named secret-npmrc, which contains the base64 encoded content of the ~/.npmrc file.

  2. Add the secret to the source section in the existing BuildConfig:

    source:
      git:
        uri: https://github.com/openshift/nodejs-ex.git
      secrets:
        - secret:
            name: secret-npmrc

To include the secret in a new BuildConfig, run the following command:

$ oc new-build \
    openshift/nodejs-010-centos7~https://github.com/openshift/nodejs-ex.git \
    --build-secret secret-npmrc

During the build, the .npmrc file is copied into the directory where the source code is located. In OpenShift Container Platform S2I builder images, this is the image working directory, which is set using the WORKDIR instruction in the Dockerfile. If you want to specify another directory, add a destinationDir to the secret definition:

source:
  git:
    uri: https://github.com/openshift/nodejs-ex.git
  secrets:
    - secret:
        name: secret-npmrc
      destinationDir: /etc

You can also specify the destination directory when creating a new BuildConfig:

$ oc new-build \
    openshift/nodejs-010-centos7~https://github.com/openshift/nodejs-ex.git \
    --build-secret “secret-npmrc:/etc”

In both cases, the .npmrc file is added to the /etc directory of the build environment. Note that for a Docker strategy the destination directory must be a relative path.

Source-to-Image Strategy

When using a Source strategy, all defined input secrets are copied to their respective destinationDir. If you left destinationDir empty, then the secrets are placed in the working directory of the builder image.

The same rule is used when a destinationDir is a relative path; the secrets are placed in the paths that are relative to the image’s working directory. The destinationDir must exist or an error will occur. No directory paths are created during the copy process.

Currently, any files with these secrets are world-writable (have 0666 permissions) and will be truncated to size zero after executing the assemble script. This means that the secret files will exist in the resulting image, but they will be empty for security reasons.

Docker Strategy

When using a Docker strategy, you can add all defined input secrets into your container image using the ADD and COPY instructions in your Dockerfile.

If you do not specify the destinationDir for a secret, then the files will be copied into the same directory in which the Dockerfile is located. If you specify a relative path as destinationDir, then the secrets will be copied into that directory, relative to your Dockerfile location. This makes the secret files available to the Docker build operation as part of the context directory used during the build.

Example 1. Example of a Dockerfile referencing secret data
FROM centos/ruby-22-centos7

USER root
ADD ./secret-dir /secrets
COPY ./secret2 /

# Create a shell script that will output secrets when the image is run
RUN echo '#!/bin/sh' > /secret_report.sh
RUN echo '(test -f /secrets/secret1 && echo -n "secret1=" && cat /secrets/secret1)' >> /secret_report.sh
RUN echo '(test -f /secret2 && echo -n "relative-secret2=" && cat /secret2)' >> /secret_report.sh
RUN chmod 755 /secret_report.sh

CMD ["/bin/sh", "-c", "/secret_report.sh"]

Users should normally remove their input secrets from the final application image so that the secrets are not present in the container running from that image. However, the secrets will still exist in the image itself in the layer where they were added. This removal should be part of the Dockerfile itself.

Custom Strategy

When using a Custom strategy, all the defined input secrets are available inside the builder container in the /var/run/secrets/openshift.io/build directory. The custom build image is responsible for using these secrets appropriately. The Custom strategy also allows secrets to be defined as described in Custom Strategy Options.

There is no technical difference between existing strategy secrets and the input secrets. However, your builder image might distinguish between them and use them differently, based on your build use case.

The input secrets are always mounted into the /var/run/secrets/openshift.io/build directory or your builder can parse the $BUILD environment variable, which includes the full build object.

Using External Artifacts

It is not recommended to store binary files in a source repository. Therefore, you may find it necessary to define a build which pulls additional files (such as Java .jar dependencies) during the build process. How this is done depends on the build strategy you are using.

For a Source build strategy, you must put appropriate shell commands into the assemble script:

.s2i/bin/assemble File
#!/bin/sh
APP_VERSION=1.0
wget http://repository.example.com/app/app-$APP_VERSION.jar -O app.jar
.s2i/bin/run File
#!/bin/sh
exec java -jar app.jar

For more information on how to control which assemble and run script is used by a Source build, see Overriding Builder Image Scripts.

For a Docker build strategy, you must modify the Dockerfile and invoke shell commands with the RUN instruction:

Excerpt of Dockerfile
FROM jboss/base-jdk:8

ENV APP_VERSION 1.0
RUN wget http://repository.example.com/app/app-$APP_VERSION.jar -O app.jar

EXPOSE 8080
CMD [ "java", "-jar", "app.jar" ]

In practice, you may want to use an environment variable for the file location so that the specific file to be downloaded can be customized using an environment variable defined on the BuildConfig, rather than updating the Dockerfile or assemble script.

You can choose between different methods of defining environment variables:

Using Docker Credentials for Private Registries

You can supply builds with a .docker/config.json file with valid credentials for private Docker registries. This allows you to push the output image into a private Docker registry or pull a builder image from the private Docker registry that requires authentication.

For the OpenShift Container Platform Docker registry, this is not required because secrets are generated automatically for you by OpenShift Container Platform.

The .docker/config.json file is found in your home directory by default and has the following format:

auths:
  https://index.docker.io/v1/: (1)
    auth: "YWRfbGzhcGU6R2labnRib21ifTE=" (2)
    email: "user@example.com" (3)
1 URL of the registry.
2 Encrypted password.
3 Email address for the login.

You can define multiple Docker registry entries in this file. Alternatively, you can also add authentication entries to this file by running the docker login command. The file will be created if it does not exist.

Kubernetes provides Secret objects, which can be used to store configuration and passwords.

  1. Create the secret from your local .docker/config.json file:

    $ oc create secret generic dockerhub \
        --from-file=.dockerconfigjson=<path/to/.docker/config.json> \
        --type=kubernetes.io/dockerconfigjson

    This generates a JSON specification of the secret named dockerhub and creates the object.

  2. Once the secret is created, add it to the builder service account. Each build is run with the builder role, so you must give it access your secret with the following command:

    $ oc secrets link builder dockerhub
  3. Add a pushSecret field into the output section of the BuildConfig and set it to the name of the secret that you created, which in the above example is dockerhub:

    spec:
      output:
        to:
          kind: "DockerImage"
          name: "private.registry.com/org/private-image:latest"
        pushSecret:
          name: "dockerhub"

    You can also use the oc set build-secret command to set the push secret on the build configuration:

    $ oc set build-secret --push bc/sample-build dockerhub
  4. Pull the builder container image from a private Docker registry by specifying the pullSecret field, which is part of the build strategy definition:

    strategy:
      sourceStrategy:
        from:
          kind: "DockerImage"
          name: "docker.io/user/private_repository"
        pullSecret:
          name: "dockerhub"

    You can also use the oc set build-secret command to set the pull secret on the build configuration:

    $ oc set build-secret --pull bc/sample-build dockerhub

This example uses pullSecret in a Source build, but it is also applicable in Docker and Custom builds.