Composition step

Run a Docker container with its dependencies inside a pipeline

The composition step runs a Docker Composition as a means to execute finite commands in a more complex interaction of services.

TIP
While composition steps are still supported, the recommended way to run integrations tests going forward is with service containers.

Motivation for Compositions

The primary purpose of compositions is to run tests that require multiple services for their execution (often known as integration tests).

The syntax offered by Codefresh closely follows the syntax for Docker-compose files, but is technically not 100% the same (there are some important differences). However, if you are already familiar with Docker compose, you will be immediately familiar with Codefresh compositions.

NOTE
Codefresh only understands Docker compose versions 2 and 3, but not point releases such as 2.1.

The big difference between the Codefresh and Docker compose is that Codefresh is distinguishes between two kinds of services:

  • Composition Services
  • Composition Candidates

Composition Services are helper services that are needed for the tests to run. These can be a database, a queue, a cache, or the backend docker image of your application – these closely parallel the services that you might define in Docker compose.

Composition Candidates are special services that will execute the tests. Codefresh will monitor their execution and the build will fail if they do not succeed. Composition candidates are almost always Docker images that contain unit/integration tests or other kinds of tests (e.g. performance)

You need at least one composition service and one candidate for the composition step.

Usage

Here is an example of a composition step. Note that there is one composition service (PostgreSQL database, named db) and one composition candidate (tests executed with gulp)

The most important part is the command line that executes the tests: command: gulp integration_test. If it fails, then the whole composition step will fail.

codefresh.yml

step_name:
  type: composition
  title: Step Title
  description: Free text description
  working_directory: ${{a_clone_step}}
  composition:
    version: '2'
    services:
      db:
        image: postgres
  composition_candidates:
    test_service:
      image: ${{build_step}}
      command: gulp integration_test
      working_dir: /app
      environment:
        - key=value
  composition_variables:
    - key=value
  fail_fast: false
  strict_fail_fast: true
  when:
    condition:
      all:
        notFeatureBranch: 'match("${{CF_BRANCH}}", "/FB-/", true) == false'
  on_success:
    ...
  on_fail:
    ...
  on_finish:
    ...
  retry:
    ...  

Caveats on sharing a docker-compose.yml

Although Codefresh’s composition syntax closely follows the syntax used in docker-compose.yml files, it is not 100% the same. If you are using docker-compose.yml locally, you may experience some problems if you try to have Codefresh reference the file (by passing it as an argument to compose, e.g. compose: docker-compose.yml).

One subtle difference is that Docker compose will interpolate environment variables that are quoted in single-braces, e.g. ${DATABASE_URL}, whereas Codefresh interpolates variables that are quoted in double-braces, e.g. ${{DATABASE_URL}}. So if your docker-compose.yml file relies on the parsing of ENV variables, it may not be a good candidate for sharing with Codefresh.

Fields

The following describes the fields available in a step of type composition

Field Description Required/Optional/Default
title The free-text display name of the step. Optional
description A basic, free-text description of the step. Optional
stage Parent group of this step. See using stages for more information. Optional
working_directory The directory in which to search for the composition file. It can be an explicit path in the container’s file system, or a variable that references another step. The default is ${{main_clone}}. Note that this is completely different from working_dir which is on the service level. Default
composition The composition you want to run. This can be an inline YAML definition or a path to a composition file on the file system, e.g. docker-compose.yml, or the logical name of a composition stored in the Codefresh system. We support most features of Docker compose version 2.0 and 3.0 Required
version Version for docker compose. Use 2 or 3 Required
composition_candidates The definition of the service to monitor. Each candidate has a single command parameter that decides what will be tested. Required
environment (service level) environment that will be accessible to the container Optional
working_dir (service level) defines the working directory that will be used in a service before running a command. By default it is defined by the docker image that is used by the service. Optional
registry_contexts Advanced property for resolving Docker images when working with multiple registries with the same domain Optional
volumes (service level) Extra volumes for individual services. Used for transferring information between your steps. Explained in detail later in this page. Optional
composition_variables A set of environment variables to substitute in the composition. Notice that these variables are docker-compose variables and NOT environment variables Optional
timeout The maximum duration permitted to complete step execution in seconds (s), minutes (m), or hours (h), after which to automatically terminate step execution. For example, timeout: 1.5h.
The timeout supports integers and floating numbers, and can be set to a maximum of 2147483647ms (approximately 24.8 days).

If defined and set to either 0s/m/h or null, the timeout is ignored and step execution is not terminated.
See Add a timeout to terminate step execution.
Optional
fail_fast Determines pipeline execution behavior in case of step failure.
  • true: The default, terminates pipeline execution upon step failure. The Build status returns `Failed to execute`.
  • false: Continues pipeline execution upon step failure. The Build status returns Build completed successfully.
    To change the Build status, set strict_fail_fast to true.
Optional
strict_fail_fast Specifies how to report the Build status when fail_fast is set to false.
You can set the Build status reporting behavior at the root-level or at the step-level for the pipeline.
  • true:
    • When set at the root-level, returns a Build status of failed when any step in the pipeline with fail_fast=false fails to execute.
    • When set at the step-level, returns a Build status of failed when any step in the pipeline with fail_fast=false and strict_fail_fast=true fails to execute.
  • false:
    • When set at the root-level, returns a Build status of successful when any step in the pipeline with fail_fast=false fails to execute.
    • When set at the step-level, returns a Build status of successful when any step in the pipeline with fail_fast=false fails to execute.

NOTES:
strict_fail_fast does not impact the Build status reported for parallel steps with fail_fast enabled. Even if a child step fails, the parallel step itself is considered successful. See also Handling error conditions in a pipeline.
Optional
when Define a set of conditions which need to be satisfied in order to execute this step.
You can find more information in the conditional execution of steps article.
Optional
on_success, on_fail and on_finish Define operations to perform upon step completion using a set of predefined post-step operations. Optional
retry Define retry behavior as described in retrying a step. Optional

Composition versus Composition Candidates

For Codefresh to determine if the step and operations were successfully executed, you must specify at least one composition_candidate.

A composition_candidate is a single service component of the normal Docker composition that is monitored for a successful exit code and determines the outcome of the step. During runtime, the composition_candidate is merged into the specified compositionand is monitored for successful execution.

The critical part of each candidate is the command parameter. This takes a single command that will be executed inside the Docker container of the candidate and will decide if the whole composition is successful or not. Only one command is allowed (similar to Docker compose). If you wish to test multiple commands you need to connect them with && like this.

 composition_candidates:
  my_unit_tests:
    image: node
    command: bash -c "sleep 60 && pwd && npm run test"

Working directories in a composition

By default, all services that take part in a composition will use as working directory the one defined by the respective image. If you want to change that, you need to use the working_dir parameter at the service level.

Here is an example:

codefresh.yml

version: '1.0'
steps:
  my_composition:
    type: composition
    title: Sample composition
    composition:
      version: '2'
      services:
        my_service:
          image: alpine
          command: 'pwd'
          working_dir: /tmp
    composition_candidates:
      my_test_service:
        image: python
        working_dir: /root
        command: 'pwd'

If you run this composition, you will see in the logs that the alpine image will use /tmp as a working directory and the python one will use /root

my_service_1       | /tmp
my_test_service_1  | /root

Add a timeout to terminate step execution

To prevent steps from running beyond a specific duration if so required, you can add the timeout flag to the step.
When defined:

  • The timeout is activated at the beginning of the step, before the step pulls images.
  • When the step’s execution duration exceeds the duration defined for the timeout, the step is automatically terminated.

NOTE
To define timeouts for parallel steps, see Adding timeouts for parallel steps.

Here’s an example of the timeout field in the step:

codefresh.yml

step_name:
  type: composition
  title: Step Title
  description: Free text description
  working_directory: ${{a_clone_step}}
  composition:
    version: '2'
    services:
      db:
        image: postgres
  composition_candidates:
    test_service:
      image: ${{build_step}}
      command: gulp integration_test
      working_dir: /app
      environment:
        - key=value
  composition_variables:
    - key=value
  timeout: 45m
  fail_fast: false
  when:
    condition:
      all:
        notFeatureBranch: 'match("${{CF_BRANCH}}", "/FB-/", true) == false'
  on_success:
    ...
  on_fail:
    ...
  on_finish:
    ...
  retry:
    ...  
Timeout info in logs

Timeout information is displayed in the logs, as in the example below.

Step termination due to timeout in logs

Step termination due to timeout in logs

Composition networking

The networking in Codefresh compositions works just like normal Docker-compose. Each service is assigned a hostname that matches its name and is accessible by other services.

Here is an example

codefresh.yml

version: '1.0'
steps:
  build_step:
    type: build
    image_name: my-node-app
    dockerfile: Dockerfile
    tag: ${{CF_BRANCH}}
  my_db_tests:
    type: composition
    composition:
        version: '2'
        services:
          db:
            image: mysql:latest
            ports:
              - 3306
            environment:
              MYSQL_ROOT_PASSWORD: admin
              MYSQL_USER: my_user
              MYSQL_PASSWORD: admin
              MYSQL_DATABASE: nodejs
    composition_candidates:
        test:
          image: ${{build_step}}
          links:
            - db
          command: bash -c 'sleep 30 && MYSQL_ROOT_PASSWORD=admin MYSQL_USER=my_user MYSQL_HOST=db MYSQL_PASSWORD=admin MYSQL_DATABASE=nodejs npm test'

In this composition the MySql instance will be available at host db:3306 accessible from the node image. When the node tests run, they will be pointed to that host and port combination to access it.

Notice also that like docker compose the order that the services are launched is not guaranteed. A quick way to solve this issue is with a sleep statement like shown above. This will make sure that the database is truly up before the tests run.

A better approach would be to use solutions such as wait-for-it which are much more robust. Here is an example:

codefresh.yml

version: '1.0'
steps:
  build_image:
    type: build
    description: Building the image...
    image_name: my-spring-boot-app
    tag: ${{CF_BRANCH_TAG_NORMALIZED}}
  build_image_with_tests:
    type: build
    description: Building the Test image...
    image_name: maven-integration-tests
    dockerfile: Dockerfile.testing
  integration_tests:
    type: composition
    title: Launching QA environment
    description: Temporary test environment
    composition:
      version: '2'
      services:
        app:
          image: ${{build_image}}
          ports:
           - 8080
    composition_candidates:
      test_service: 
        image: ${{build_image_with_tests}}
        links:
          - app
        command: bash -c '/usr/bin/wait-for-it.sh -t 20 app:8080 -- mvn verify -Dserver.host=app'

In this composition a Java application is launched at app:8080 and then a second image is used for integration tests that target that URL (passed as a parameter to Maven).

The wait-for-it.sh script will make sure that the Java application is truly up before the tests are started. Notice that in the example above the script is included in the testing image (created by Dockerfile.testing)

Using public Docker images in a composition

It is important to notice that Docker images used in a composition (both as services and candidates) will be looked from your connected registries first before looking at Dockerhub:

codefresh.yml

version: "1.0"
steps:
  my_composition:
    type: composition
    title: Sample composition
    composition:
      version: '2'
      services:
        my_service:
          image: mysql
          ports:
            - 3306
    composition_candidates:
      my_test_service:
        image: alpine
        working_dir: /root
        command: 'pwd'

In the example above if you already have two images in your private registries named mysql and alpine, then THEY will be used instead of the respective images in Dockerhub.

You can see which images are used in the logs of the builds:

Running composition step: Sample composition                                                                                              
Pulling kostisazureregistry.azurecr.io/mysql@sha256:1ee5515fed3dae4f13d0f7320e600a38522fd7e510b225e68421e1f90                      
Pulling kostisazureregistry.azurecr.io/alpine@sha256:eddb7866364ec96861a7eb83ae7977b3efb98e8e978c1c9277262d327                     

Accessing your project folder from a composition

By default, the services of a composition run in a completely isolated manner. There are several scenarios however where you wish to access your Git files such as:

  • Using test data that is available in the project folder
  • Preloading a database with a data script found in Git
  • Running integration tests and then using their results for reporting

The Codefresh shared volume is automatically mounted in freestyle steps but NOT in compositions. You have to mount it yourself if you use that functionality.

Here is an example where the shared volume is mounted in a composition – '${{CF_VOLUME_NAME}}:${{CF_VOLUME_PATH}}' is listed under volumes:

codefresh.yml

version: '1.0'
steps:
  create_test_data_step: 
    title: Creating dummy data
    image: alpine
    commands:
      - echo "Writing in shared volume" > /codefresh/volume/sample_text.txt 
  my_sample_composition:
    type: composition
    title: Composition with volume
    composition:
      version: '2'
      services:
        my_sample_service:
          image: node
          volumes:
            - '${{CF_VOLUME_NAME}}:${{CF_VOLUME_PATH}}'
          working_dir: '${{CF_VOLUME_PATH}}'
          command: bash -c "pwd && cat sample_text.txt"
    composition_candidates:
      my_unit_tests:
        image: python
        volumes:
          - '${{CF_VOLUME_NAME}}:${{CF_VOLUME_PATH}}'
        working_dir: '${{CF_VOLUME_PATH}}'
        command: bash -c "pwd && echo 'Finished tests' > test_result.txt"
  read_test_data_step:
    title: Reading dummy data
    image: alpine
    commands:
      - ls -l /codefresh/volume
      - cat /codefresh/volume/test_result.txt 

In this pipeline:

  1. The first freestyle step writes a simple test file in the shared volume.
  2. The composition starts and both services (my_sample_service and my_unit_tests) attach the same volume.
  3. The sample service reads from the shared volume (i.e. using test data that was created before).
  4. The sample unit test service writes to the shared volume (emulating test results).
  5. The last freestyle step reads the file that was written by the composition.

Therefore, in this pipeline you can see both ways of data sharing, bringing files into a composition and getting results out of it. Notice that we need to mount the shared volume only in the composition services. The freestyle steps automatically mount /codefresh/volume on their own.

NOTE
To mount the shared volume in one of your composition services, you must mount it in the composition_candidate also. It is not compulsory to mount the shared volume in all services of a composition. Only those that actually use it for file transfer, should mount it.

Composition variables versus environment variables

Docker compose supports two kinds of variables in its syntax :

  • There are environment variables that are used in the docker-compose file itself (${VAR} syntax).
  • There are environment variables that are passed in containers (environment: yaml group).

Codefresh supports both kinds, but notice that variables mentioned in the composition_variables yaml group refer to the first kind. Any variables defined there are NOT passed automatically to containers (use the environment yaml group for that purpose).

This can be illustrated with the following example:

codefresh.yml

version: '1.0'
steps:
  comp1:
    type: composition
    title: Composition example 1
    description: Free text description
    composition:
      version: '2'
      services:
        db:
          image: alpine
    composition_candidates:
      test_service:
        image: alpine
        command: printenv
        environment:
          - FIRST_KEY=VALUE
    composition_variables:
      - ANOTHER_KEY=ANOTHER_VALUE

If you run the compositio,n you will see that the printenv command shows the following:

test_service_1  | FIRST_KEY=VALUE

The FIRST_KEY variable which is defined explicitly in the environment yaml part is correctly passed to the alpine container. The ANOTHER_KEY is not visible in the container at all.

You should use the composition_variables yaml group for variables that you wish to reuse in other parts of your composition using the ${ANOTHER_KEY} syntax.

Merging services

If the composition already contains a service with the same name as the composition_candidate, the two service definitions are combined, with preference given to the composition_candidate’s definition.

For example, we create a new Codefresh composition named ‘test_composition’:

test-composition.yml

version: '2'
  services:
    db:
      image: postgres
    test_service:
      image: myuser/mytestservice:latest
      command: gulp integration_test

Now we want to reuse this composition during our build for testing purposes. We can add the following composition step to our codefresh.yml file and define the composition step so that test_service always uses the latest image that was built.

YAML

run_tests:
  type: composition
  composition: test_composition
  composition_candidates:
    test_service:
      image: ${{build_step}}

In the above example, both composition and composition_candidates define a service named test_service. After merging these definitions, test_service will maintain the command that was defined in the original composition but will refer to the image built by the step named build_step.

Steps in pipelines
Variables in pipelines
Introduction to pipelines
Integration testing