Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
452 lines (358 sloc) 16.8 KB

Pipelines

Concourse is an integral part of Genesis v2. But rather than force operators to define the full structure of a Concourse deployment pipeline (which can get rather complex), we want to provide them a domain-specific language for specifying their pipelines.

auto sandbox-*
auto preprod-*

sandbox-a -> sandbox-b
sandbox-b -> preprod-a
preprod-a -> preprod-b

preprod-b -> prod-a
prod-a    -> prod-b

We want to stay away from YAML for this one for a few reasons: (a) to avoid the confusion of what gets merged with what, (b) to constrain the overall pipeline structure to a single, small file, and (c) I haven't found a good YAML structure for specifying the relationships that doesn't hurt my brain.

To wit, the pipeline language will need to be able to:

  • Identify the structure of the pipeline, explicitly and unambiguously.
  • Configure notifications for non-triggered deployments.
  • Configure the name of the pipeline.
  • Handle incomplete subsets of the deployment set.
  • Configure the Concourse endpoint for update operations.

Propagation

The purpose of a pipeline is to test and vet configuration in less important environments (sandbox, staging, pre-production) before attempting to deploy them to more important environments (prod).

This can be handled directly by operations staff by manually controlling which environments get changes and porting configuration changes between the environments when deployments succeed, but we can do better.

Consider the following files:

  • foobar.yml
  • foobar-aws.yml
  • foobar-aws-us1.yml
  • foobar-aws-us1-preprod.yml
  • foobar-aws-us1-preprod-east.yml *
  • foobar-aws-us1-preprod-west.yml *
  • foobar-aws-us1-prod.yml
  • foobar-aws-us1-prod-east.yml *
  • foobar-aws-us1-prod-west.yml *
  • foobar-aws-eu1.yml
  • foobar-aws-eu1-prod.yml
  • foobar-aws-eu1-prod-east.yml *
  • foobar-aws-eu1-prod-west.yml *

That's a lot of files, and not all of them need to exist, but we do need to nail down a specific and well-defined set of rules for dealing with change at various levels.

(Keep in mind that what used to be site/ lives in cloud-config and global/ is in our Genesis kits)

Now let's assume that we've configured our pipeline as follows:

foobar-aws-us1-preprod-east  ->   foobar-aws-us1-preprod-west
foobar-aws-us1-preprod-west  ->   foobar-aws-us1-prod-east
foobar-aws-us1-preprod-west  ->   foobar-aws-eu1-prod-east

foobar-aws-us1-prod-east     ->   foobar-aws-us1-prod-west

foobar-aws-eu1-prod-east     ->   foobar-aws-eu1-prod-west

Or, put more succintly, deploy east followed by west, and the U.S. preprod environments lead to deployment of both U.S. and E.U. production.

Let's look at the foobar-aws-us1-preprod-west, which should trigger after the east side. When it does, Genesis needs to merge the following files:

  • foobar.yml
  • foobar-aws.yml
  • foobar-aws-us1.yml
  • foobar-aws-us1-preprod.yml
  • foobar-aws-us1-preprod-west.yml

All but the last file are shared with the east environment, and have been vetted by the successful runs of the pipeline. That means we can and should use that version, regardless of what has changed in the interim.

If that last sentence seems confusing, consider the effect that time may have on the world. Deployment in the east side could succeed on Monday, but be delayed (either through failure, or because of required manual intervation) in the west. On Thursday, when the west side is deployed again, we want to explicitly ensure that any intervening changes to common files are not included, since those changes have not been vetted in the east.

This is where Genesis caching comes into play.

After a successful deployment of the east side, all of the intermediate files used to generate that manifest will be cached in a special directory under .genesis/. Pipelines will be configured to watch the cached files shared by any given environment and its immediate predecessor environment. In this case, that would be:

  • foobar.yml
  • foobar-aws.yml
  • foobar-aws-us1.yml
  • foobar-aws-us1-preprod.yml

foobar-aws-us1-preprod-west.yml of course, would be pulled from the main area, as it cannot (by definition) be cached in any other environment.

To explore this method of caching, I've written pipey, which you can find in the bin/ directory of this repository. Pipey takes two environment names (from and to) and figures out which hypothetical files it should watch in from's cache, and which it should include directly.

Here are some example runs.

East-to-West propagation within the same region and environment:

pipeline progression
  from [foobar-aws-us1-preprod-east]
    to [foobar-aws-us1-preprod-west]

CACHED foobar.yml                       (from .genesis/cached/foobar-aws-us1-preprod-east/foobar.yml)
CACHED foobar-aws.yml                   (from .genesis/cached/foobar-aws-us1-preprod-east/foobar-aws.yml)
CACHED foobar-aws-us1.yml               (from .genesis/cached/foobar-aws-us1-preprod-east/foobar-aws-us1.yml)
CACHED foobar-aws-us1-preprod.yml       (from .genesis/cached/foobar-aws-us1-preprod-east/foobar-aws-us1-preprod.yml)
DIRECT foobar-aws-us1-preprod-west.yml

Triggering prod in the same region:

pipeline progression
  from [foobar-aws-us1-preprod-west]
    to [foobar-aws-us1-prod-east]

CACHED foobar.yml                       (from .genesis/cached/foobar-aws-us1-preprod-west/foobar.yml)
CACHED foobar-aws.yml                   (from .genesis/cached/foobar-aws-us1-preprod-west/foobar-aws.yml)
CACHED foobar-aws-us1.yml               (from .genesis/cached/foobar-aws-us1-preprod-west/foobar-aws-us1.yml)
DIRECT foobar-aws-us1-prod.yml
DIRECT foobar-aws-us1-prod-east.yml

Triggering prod in a different region:

pipeline progression
  from [foobar-aws-us1-preprod-west]
    to [foobar-aws-eu1-prod-east]

CACHED foobar.yml                       (from .genesis/cached/foobar-aws-us1-preprod-west/foobar.yml)
CACHED foobar-aws.yml                   (from .genesis/cached/foobar-aws-us1-preprod-west/foobar-aws.yml)
DIRECT foobar-aws-eu1.yml
DIRECT foobar-aws-eu1-prod.yml
DIRECT foobar-aws-eu1-prod-east.yml

Configuration

There is a temptation to unify the BOSH manifest configuration with the Concourse Pipeline manifest. The authors fell victim to this temptation on multiple occasions. We must, however, resist as best we can, for a few key reasons:

First, generating a pipeline configuration necessarily involves all environments being pipelined. If the configuration for the pipeline (i.e. BOSH credentials, worker tags, etc.) is spread out across multiple files, Genesis will have no recourse but to generate each and every manifest to put the configuration "back together" and create the Concourse pipeline file.

This is problematic on two fronts. It prolongs the process of pipeline configuration, due to the additional work of parsing, merging and evaluating all of those manifest files. More severely, it places undue restrictions on the Vault architecture chosen by the operations team.

Consider what happens when you have multiple disparate virtual private clouds, and for security reasons, you wish to run a dedicated Vault in each, such that Vault traffic need not transit the public Internet. If you attempt to deploy a single unified pipeline (perhaps utilizing remote Concourse workers), you quickly run into the fact that you can't actually generate each environments manifest from outside of that environment. How do you get to the Vault?

Instead, we should try to unify as much of the pipeline configuration into a single, Spruce-merged configuration file. We think it looks a little something like this:

pipeline:
  name: redis-deployments
  git:
    owner: someco
    repo:  something-deployments
    private_key: (( vault "secret/concourse/git:private" ))

  slack:
    channel: '#botspam'
    webhook: (( vault "secret/concourse/slack:webhook" ))

  email:
    stanza: here

  vault:
    url:    https://127.0.0.1:8200
    verify: yes

  stemcells:
    bosh-lite: bosh-warden-boshlite-ubuntu-trusty-go_agent
    aws: bosh-aws-xen-hvm-ubuntu-trusty-go_agent

  boshes:
    sandbox:
      alias:    sb # Optional
      url:      https://sandbox.example.com:25555
      ca_cert:  (( vault "secret/bosh/sandbox/ssl/ca:certificate" ))
      username: sb-admin
      password: (( vault "secret/bosh/sandbox/admin:password" ))
      stemcells:
      - bosh-lite

    preprod:
      url:      https://preprod.example.com:25555
      ca_cert:  (( vault "secret/bosh/preprod/ssl/ca:certificate" ))
      username: pp-admin
      password: (( vault "secret/bosh/preprod/admin:password" ))
      stemcells:
      - aws

    prod:
      url:      https://prod.example.com:25555
      ca_cert:  (( vault "secret/bosh/prod/ssl/ca:certificate" ))
      username: pr-admin
      password: (( vault "secret/bosh/prod/admin:password" ))
      stemcells:
      - aws

  smoke-tests: run-my-smoke-tests
  tagged: yes

  layouts:
    # genesis repipe                 ; if target is 'default'
    # genesis repipe -t azure        ; if target is 'azure' instead
    default: |+
      auto *sandbox *preprod
      sandbox -> preprod -> prod

    # genesis repipe onprem          ; if target is 'onprem'
    # genesis repipe -t ci onprem    ; if it is 'ci' instead
    onprem: |+
      auto *sandbox *preprod
      on-prem-1-sandbox -> on-prem-1-preprod -> on-prem-1-prod
      on-prem-2-sandbox -> on-prem-2-preprod -> on-prem-2-prod

Pipeline YAML Field Guide

  • pipeline.name - The name of your pipeline, as displayed in Concourse (i.e. 'redis-deployments'). This setting is required.

  • pipeline.public - Whether or not the pipeline is visible to everyone. Setting this to yes will make the Concourse pipeline (and all logs related to deployments) public. This is off by default.

  • pipeline.tagged - Whether or not the Concourse pipeline configuration uses tags on each job, for use in multi-site, distributed Concourse implementations. This is off by default, and if you don't know why you would need it, you don't need it.

  • pipeline.unredacted - Whether or not deployment output will be redacted when run in the pipeline. This is off by default. If you set it to 'yes', the genesis deploy output in pipeline build logs will be unredacted, showing you all of the changes that are taking place, including any changes in potentially sensitive credentials.

  • pipeline.smoke-tests - The name of the BOSH smoke test errand to run after a successsful deployment / upgrade. If not specified, no smoke testing will be carried out.

  • pipeline.skip_upkeep - If set, disables the automatic stemcell updating behavior of the pipeline.

  • pipeline.stemcells - A map of stemcell alias to stemcell name. This configures what stemcells are tracking new releases. The alias is used for abbreviating stemcell name display in the Concourse UI, as well as in ci.yml. The name, must match an official BOSH stemcell name found on https://bosh.io/stemcells.

  • pipeline.vault.url - The URL of your Vault installation, i.e. https://vault.example.com. This is required.

  • pipeline.vault.role - The AppRole GUID of a given Vault AppRole, used to generate temporary tokens for Vault-accessing during deploys. This field is optional, provided that the setup-approle Genesis addon was executed on the targeted Concourse. If you'd prefer to manage your own AppRole and policy, you may fill out this field.

  • pipeline.vault.secret - The secret key GUID of a given Vault AppRole, used to authenticate the AppRole to vault. Like the above field, it is optional provided that the setup-approle addon was executed on the targeted Concourse. If you'd prefer to manage your own AppRole and policy, you may fill out this field.

  • pipeline.vault.verify - Instruct Concourse to validate the Vault TLS certificate (if using https:// for your Vault). This is on by default, and you probably don't want to change it.

  • pipeline.git.host - The hostname of Github. Useful for on-premise, Enterprise Github installations. Defaults to github.com.

  • pipeline.git.owner - The name of the user or organization who owns the Genesis deployment repository, in Github. This is required.

  • pipeline.git.repo - The name of the Genesis deployment repository, in Github. This is required.

  • pipeline.git.private_key - The private component (i.e. ----BEGIN RSA PRIVATE KEY--- ...) of the SSH Deployment Key for this repository, in Github. Since the pipeline will push to the repository, this key needs to be installed in the Github web interface with the Allow write access box checked.

  • pipeline.slack.webhook - The Slack Integration WebHook URL, provided by Slack when you configure a new integration. This is required.

  • pipeline.slack.channel - The name of the channel (#channel) or user (@user) who will receive Slack notifications. This is required if you wish to use Slack for notifications.

  • pipeline.slack.username - The username to use when posting notifications to Slack. Defaults to runwaybot.

  • pipeline.slack.icon - URL of an image to use as the avatar / profile picture when posting notifications to Slack. Defaults to an airplane-looking thing.

  • pipeline.stride.client_id - The Client ID of the Stride app that you would like to post notifications as. required.

  • pipeline.stride.client_secret - The Client Secret of the Stride app that you would like to post notifications as. required.

  • pipeline.stride.cloud_id - The ID of the Stride cloud that you would like to post to. required.

  • pipeline.stride.conversation - Name of the stride conversation (channel) that you would like notifications to go to. required

  • pipeline.email.to - A list of email addresses to send notifications to. This is required if you wish to use email for notification.

  • pipeline.email.from - An email address from which to send notification emails. This is required.

  • pipeline.email.smtp.host - The IP address or FQDN of the SMTP relay / mail server to send email through. This is required.

  • pipeline.email.smtp.port - The SMTP port that the relay / mail server listens on for mail submission. Defaults to the standard mail submission port, 587 (not 25).

  • pipeline.email.smtp.username - The username to authenticate to the SMTP relay / mail server as. This is required. Anonymous relays are not supported, as they are harmful to the Internet.

  • pipeline.email.smtp.password - The password to use when authenticating to the SMTP relay / mail server. This is required.

  • pipeline.boshes.<env-name>.url - The URL to the BOSH director API (i.e. what you would normally bosh target). Each environment to be deployed via the pipeline must have one of these set.

  • pipeline.boshes.<env-name>.alias - An optional alias to use in the concourse pipeline, when generating job names, and resource names, in case your bosh env-name is rather long. If not specified, the env-name will be used.

  • pipeline.boshes.<env-name>.ca_cert - The CA Certificate of the BOSH director. This is required.

  • pipeline.boshes.<env-name>.username - The username to use when authenticating to this BOSH director. This is required.

  • pipeline.boshes.<env-name>.password - The password for the configured BOSH user account. This is required.

  • pipeline.boshes.<env-name>.stemcells - A list of stemcell aliases that this environment will require. Genesis will automatically upload new versions of these stemcells as they come out, triggering new deployments, or notify you of the pending changes, in the case of manual deployments. This list must be present, unless pipeline.skip_upkeep is set. If you wish to have the pipeline manage stemcells for one environment, but not another, you may set this to an empty array.

  • pipeline.task.image - The name of the Docker image to use for running tasks. This defaults to starkandwayne/concourse, and you are highly encouraged not to change it without good reason.

  • pipeline.task.version - The version of the Docker image to use for running tasks. This defaults to latest, which should work well for most implementations.

Pipeline Configuration

If you need to edit and make changes to your deployment pipeline, simply add the changes you need to the bottom of your ci.yml located in your *-deployments repo. For example, if we have a pipeline we need to edit the 'check_every:' parameter of a resource named git we can add the block below under our pipeline layouts:

ci.yml

layouts:
  test-sandbox-ops: |+
    auto *test-sandbox *test-staging
    test-sandbox -> test-staging

resources:
- name: git
  check_every: 15m
You can’t perform that action at this time.