pool-resource

a pool of locks (modeling semaphores)

Allows you to lock environments, pipeline flow, or other entities which have to be interacted with in a serial manner.

This resource is backed by a Git repository, so the configuration is largely the same as the git-resource.

Additional features compared to the upstream pool resource

This fork provides the following additional features:

1. The commit message on each lock operation contains the team name. This enables the smart reaping of stale locks in concert with the Pool Boy script.
2. The resource is instrumented to report metrics to a Prometheus Pushgateway. This allows to gather important information such as queue length (number of jobs waiting on a lock). Usage is explained in this document.

Git Repository Structure

.
├── aws
│   ├── claimed
│   │   ├── .gitkeep
│   │   └── env-2
│   └── unclaimed
│       ├── .gitkeep
│       └── env-1
├── ping-pong-tables
│   ├── claimed
│   │   └── .gitkeep
│   └── unclaimed
│       ├── .gitkeep
│       └── north-table
└── vsphere
    ├── claimed
    │   ├── .gitkeep
    │   └── f3cb3823-a45a-49e8-ab41-e43268494205
    └── unclaimed
        ├── .gitkeep
        └── 83ed9977-3a10-4c49-a818-2d7a37693da7

This structure represents 3 pools of locks, aws, ping-pong-tables, and vsphere. The .gitkeep files are required to keep the unclaimed and claimed directories track-able by Git if there are no files in them.

You will need to mirror this structure in your own lock repository. In other words, initialize an empty repository and create one directory in the root of the repository for each pool of locks (e.g. aws). Inside of each lock pool directory, create one directory named claimed and one directory named unclaimed. Inside each claimed and unclaimed directory, create an empty filed named .gitkeep. Finally, create individual locks by making an empty file inside of the unclaimed directory (assuming you want your lock to be unclaimed by default) with the desired name of the lock (e.g. env-1).

Source Configuration

• uri: Required. The location of the repository.

• branch: Required. The branch to track.

• pool: Required. The logical name of your pool of things to lock.

• private_key: Optional. Private key to use when pulling/pushing. Ensure it does not require a password. Example:

  private_key: |
    -----BEGIN RSA PRIVATE KEY-----
    MIIEowIBAAKCAQEAtCS10/f7W7lkQaSgD/mVeaSOvSF9ql4hf/zfMwfVGgHWjj+W
    <Lots more text>
    DWiJL+OFeg9kawcUL6hQ8JeXPhlImG6RTUffma9+iGQyyBMCGd1l
    -----END RSA PRIVATE KEY-----

• username: Optional. Username for HTTP(S) auth when pulling/pushing. This is needed when only HTTP/HTTPS protocol for git is available (which does not support private key auth) and auth is required.

• password: Optional. Password for HTTP(S) auth when pulling/pushing.

• retry_delay: Optional. If specified, dictates how long to wait until retrying to acquire a lock or release a lock. The default is 10 seconds. Example of valid values: 60s, 90m, 1h.

• prometheus_push_gateway: Optional. URL to the Prometheus push gateway that will expose the following metrics for each execution of the out step:

  • resource_pool_execution_time: Total execution time of the current lock operation. Labels: pool, branch, operation
  • resource_pool_num_available_locks: Number of available/unclaimed locks in the current pool. Labels: pool, branch
  • resource_pool_num_claimed_locks: Number of claimed locks in the current pool. Labels: pool, branch

  Note: All metrics are computed after performing the lock operation (acquire, release, etc). There is no uri label, because Prometheus cannot handle slashes (/) or many other common characters present in a git URI. Instead, you may consider relying on either unique pool names, and/or branch names depending on the context. For example, you could restrict branch: master for production and branch: stage for smoke tests.

Behavior

check: Check for changes to the pool.

The repository is cloned (or pulled if already present), and any commits made to the specified pool from the given version on are returned. If no version is given, the ref for HEAD is returned.

in: Fetch an acquired lock.

Outputs 2 files:

• metadata: Contains the contents of whatever was in your lock file. This is useful for environment configuration settings.

• name: Contains the name of lock that was acquired.

out: Acquire, release, add, or remove a lock.

Performs one of the following actions to change the state of the pool.

Parameters

One of the following is required.

• acquire: If true, we will attempt to move a randomly chosen lock from the pool's unclaimed directory to the claimed directory. Acquiring will retry until a lock becomes available.

• claim: If set, the specified lock from the pool will be acquired, rather than a random one (as in acquire). Like acquire, claiming will retry until the specific lock becomes available.

• release: If set, we will release the lock by moving it from claimed to unclaimed. The value is the path of the lock to release (a directory containing name and metadata), which typically is just the step that provided the lock (either a get to pass one along or a put to acquire).

  Note: the lock must be available in your job before you can release it. In other words, a get step to fetch metadata about the lock is necessary before a put step can release the lock.

• add: If set, we will add a new lock to the pool in the unclaimed state. The value is the path to a directory containing the files name and metadata which should contain the name of your new lock and the contents you would like in the lock, respectively.

• add_claimed: Exactly the same as the add param, but adds a lock to the pool in the claimed state.

• remove: If set, we will remove the given lock from the pool. The value is the same as release. This can be used for e.g. tearing down an environment, or moving a lock between pools by using add with a different pool in a second step.

• update: If set, we will update an existing lock in the pool.

  • If the existing lock is in the unclaimed state we will update it with the contents of the metadata file.
  • If no such lock is present in either the claimed or unclaimed state we add a new lock to the pool in the unclaimed state.
  • If the lock is in the claimed state we will wait for it to be unclaimed and proceed to update it as above.

  The value is the path to a directory containing the files name and metadata which should contain the name of your new lock and the contents you would like in the lock, respectively.

Note: all out operations described above will generate a git commit. This fork of the original Concourse pool resource also describes, in the git commit message, the concourse team name that ran the out step.

Example Concourse Configuration

The following example pipeline models acquiring, passing through, and releasing a lock based on the example git repository structure:

resources:
- name: aws-environments
  type: pool
  source:
    uri: [email protected]:concourse/locks.git
    branch: master
    pool: aws
    private_key: |
      -----BEGIN RSA PRIVATE KEY-----
      MIIEowIBAAKCAQEAtCS10/f7W7lkQaSgD/mVeaSOvSF9ql4hf/zfMwfVGgHWjj+W
      <Lots more text>
      DWiJL+OFeg9kawcUL6hQ8JeXPhlImG6RTUffma9+iGQyyBMCGd1l
      -----END RSA PRIVATE KEY-----

jobs:
- name: deploy-aws
  plan:
    - put: aws-environments
      params: {acquire: true}
    - task: deploy-aws
      file: my-scripts/deploy-aws.yml

- name: test-aws
  plan:
    - get: aws-environments
      passed: [deploy-aws]
    - task: test-aws
      file: my-scripts/test-aws.yml
    - put: aws-environments
      params: {release: aws-environments}

Managing multiple locks

The parameter for release is the name of the step whose lock to release. Normally this is just the same as the name of the resource, but if you have custom names for put (for example if you need to acquire multiple instances of the same pool), you would put that name instead. For example:

- name: test-multi-aws
  plan:
    - put: environment-1
      resource: aws-environments
      params: {acquire: true}
    - put: environment-2
      resource: aws-environments
      params: {acquire: true}
    - task: test-multi-aws
      file: my-scripts/test-multi-aws.yml
    - put: aws-environments
      params: {release: environment-1}
    - put: aws-environments
      params: {release: environment-2}

Development

Prerequisites

• golang is required - version 1.10.x is tested; earlier versions may also work.
• docker is required - version 18.03.x is tested; earlier versions may also work.
• godep, for syncing dependencies
• counterfeiter, for generating test-doubles.

Running the tests

The tests have been embedded with the Dockerfile; ensuring that the testing environment is consistent across any docker enabled platform. When the docker image builds, the test are run inside the docker container, on failure they will stop the build.

Run the tests with the following command:

docker build -t pool-resource .

Contributing

Please make all pull requests to the master branch and ensure tests pass locally.