Skip to content

Latest commit

 

History

History
417 lines (326 loc) · 14.3 KB

design.md

File metadata and controls

417 lines (326 loc) · 14.3 KB
Raw
 _           _                     _
| |__   __ _| |_   _  __ _ _ __ __| |
| '_ \ / _` | | | | |/ _` | '__/ _` |
| | | | (_| | | |_| | (_| | | | (_| |
|_| |_|\__,_|_|\__, |\__,_|_|  \__,_|
               |___/

This document is made of three key sections, and we recommend reading them in order.

Architecture

In order to understand how Halyard will tackle the problems we will describe below, it is valuable to have a mental model of how it will be architected.

Since we want to provide both a generic interface to make configuration requests against, without commiting to requiring a CLI, a web UI, a packaged library, etc... we will start by writing Halyard as a daemon, listening to requests on port 80641. This allows the content of those requests to be generated by a user-friendly interface or library which will be described in a separate document.

Goals and Solutions

Currently, it is not easy to setup and operate non-trivial installations of Spinnaker. It requires knowledge of the behavior and operation of many independently configured services - and even with that knowledge - there is no existing solution for distributing or valdating configuration, updating Spinnaker's stateful services, or interacting with Spinnaker's API outside of the prebuilt UI. The goal of Halyard is to provide a solution for all of this. To understand what Halyard will become, and how it will address these problems, we will separate the concerns of Halyard into two stages.

Since this document focuses only on the Halyard daemon, it won't be immediately clear what the value the config files documented here bring to the user, since the end goal is to simplify config after all, not create more of it. However, these files are really meant to be read and modified only by the daemon, whereas the user would modify their config with a series of commands like this:

$ hal add-account --help
You need to specify
   --provider (kubernetes, aws, or google)
   --name (a human readable name for this account)
as well as some provider specific information you will be prompted for.
$ hal add-account --provider kubernetes --name prod
The kubernetes provider requires
   --context
$ hal add-account --provider kubernetes --name prod --context <TAB><TAB>
gke_us-central-1       gke_us-east-1   # tab completed entries
gke_us-west-1          gke_us-east-2
$ hal add-account --provider kubernetes --name prod --context gke_us-central-1
Account added validated & added successfully.
$ hal add-account --provider kubernetes --name prod --context gke_us-central-3
Account "prod" could not be added:
  - Context "gke_us-central-3" is not a valid entry in ~/.kube/config.
  - Account "prod" for provider "kubernetes" already exists.
$ hal list-accounts --provider kubernetes
- prod
- staging
- test
$ hal update --service clouddriver
Updating clouddriver configuration...
Creating new clouddriver cluster...
Spinning down old clouddriver cluster...
Update successful

Previously, the update path has failed for users in the following places:

  • A) Understanding all of the configuration options availabe to the Kubernetes provider (there are at least 10 or so) after finding the correct documentation.

  • B) Understanding how to add multiple accounts to Spinnaker (this comes up weekly in the slack channel).

  • C) Realizing that they have to edit clouddriver-local.yaml rather than spinnaker-local.yaml, because only the former supports multiple accounts per-provider.

  • D) Validating that the context they have chosen actually works. This wouldn't be discovered until after clouddriver has been restarted, and the account fails to show up with an error message deep inside the logs.

These are common issues that happen with each provider, and provide a serious user experience problem when running Spinnaker. Halyard's objective is programatically overcome these issues.

1) Configuring Spinnaker

The first stage in Halyards development will involve three parts.

1.1) Versioning Spinnaker

In order to have confidence in the Spinnaker installation being deployed, we need to pin specific versions of Spinnaker microservices, as well as the dependencies they require in a Bill of Materials (BOM). We propose that the schema looks like this:

version: 1.4.0
services:                   # one entry for every service
  clouddriver:
    version: 1.320.0        # corresponds to travis-ci spinnaker release
    dependencies:           # list of name/version pairs
      - name: redis
        version: >2.0       # it is worth exploring version ranges here
  orca: ...

While the first iteration of Halyard development (configuring Spinnaker) will not be able to deploy Spinnaker with the specified versions, it is important that we pin sets of Spinnaker configuration to sets of Spinnaker service versions by means of a single version number.

This BOM should never need to exist on any machine deploying or running Spinnaker, as it only needs to be readable by Halyard at deployment/configuration time, meaning it could be hosted at a publicly readable web endpoint. However, the Spinnaker version itself will be present on whatever machine is running Halyard to inform it of what configuration to read and what to deploy.

The key takeaway here is that a Spinnaker version points to a BOM, and that Halyard therefore only needs the Spinnaker version to determine the version of all its dependencies

1.2) Distributing Authoritative Configuration

The idea is to have a single place that shared, authoritative Spinnaker configuration can be downloaded from. This will ultimately replace the configuration in spinnaker/config by storing each *.yaml file in a single versioned bucket (S3/GCS). The bucket version will be mapped to a Spinnaker version to make it simple for Halyard to fetch the correct configuration. The actual set of configuration will never need to be stored on the maching running Halyard, only staged there during distribution of the configuration. This configuration (alongside user edits) will be placed baked into the VMs running Spinnaker services.

1.3) Generating User Configuration

This will be the most challenging part of Halyard's first phase of development. In order to do this correctly, let's first list some goals:

  • a) The user should never have to open a text editor to write or edit Spinnaker confiuration

  • b) If the user does want to hand-edit configuration, Halyard should not interfere with that (but it will be an advanced use-case, and shall be treated as such).

  • c) Halyard should enable a user to configure multiple instances of Spinnaker all from the same machine.

  • d) It should be easy to extend Halyard to accept new config options.

To achieve these goals, Halyard will take a two-step approach to generating Spinnaker configuration:

  • a) Receive a number of user commands (add an account, add a trigger, etc...) and store the resulting output in the ~/.hal/config file.

  • b) Reading configuration from ~/.hal/config and from the specified Spinnaker version, write out all Spinnaker configuration to ~/.spinnaker (the default configuration directory).

Before exploring the semantics of the individual Halyard commands, let's look at the ~/.hal directory.

~/.hal Contents

The directory structure will look something like this:

.hal/
  config                     # all halyard spinnaker entries
  spinnaker-api-team/        # optional directory with per-spinnaker config
    clouddriver-local.yaml   # optional -<profile>.yaml files with overrides
  spinnaker-ml-team/
    clouddriver-local.yaml

The takeaway for the above diagram is that only ~/.hal/config is required and read by Halyard, and that for each separate installation of Spinnaker you can optionally provide your own *-<profile>.yaml files for further configuration. The things configured in these *-<profile>.yaml files should not conflict with any configured in the below ~/.hal/config.

The contents of ~/.hal/config will look like this:

halyard-version: 1.0.0
current-deployment: spinnaker-api-team # which deployment to operate on
deployment-configuration:
  - name: spinnaker-api-team
    version: 1.4.0                     # Spinnaker version
    providers: &clouddriver            # anchor referenced in clouddriver.yaml
      kubernetes:                      # provider-specific details
        enabled: true
        accounts:
          - name: my-kubernetes-account
            context: ...
          - name: my-other-kubernetes-account
      google: &google
        enabled: false
        accounts:
          - name: ...
    webhooks: &igor                    # anchor referenced in igor.yaml
      jenkins:                         # CI-specific details
        enabled: true
        accounts:
          - name: cloudbees
            address: ...
  - name: spinnaker-ml-team
    accounts: ...

The anchors referenced above (e.g. &clouddriver) there will reference entries in each canned *.yaml file like so:

# clouddriver.yaml
<<: *clouddriver        # This merges all clouddriver entries in ~/.hal/config

~/.hal Semantics

Now that we know what will be stored in the ~/.hal directory, we need to explain how to generate the contents of the ~/.spinnaker directory.

  • a) For the current deployment (selected by current-deployment in ~/.hal/config), Halyard will download all configuration for the given version number, and prepend the current-deployment's deployment-configuration into each canned configuration file (e.g. clouddriver.yaml).

  • b) Halyard will copy the entries in ~/.hal/<deployment>/*-<profile>.yaml into ~/.spinnaker.

Notice that so far we have just defined a more general spinnaker-local.yaml, which alone is not interesting. The reason this is necessary is that spinnaker-local.yaml prevented us from refering to nested dictionaries in the way the anchors in conjunction with ~/.hal/config do now. With this in place, we can fully configure Spinnaker's core options (providers, git, webhooks, etc...) in a centralized file. The ability to add additional *-<profile>.yaml files exists only to cover config options that are very unlikely to be touched by users of Spinnaker.

~/.hal Example

To drive home the interaction of the *-<profile>.yaml files, the ~/.hal/config files, and the CVM consider the following example:

We have the following directory structure:

.hal/
  config
  spinnaker-api-team/
    clouddriver-local.yaml

# ~/.hal/config
halyard-version: 1.0.0
current-deployment: spinnaker-api-team
deployment-configuration:
  - name: spinnaker-api-team
    version: 1.4.0
    providers: &clouddriver
      kubernetes:
        enabled: true
        accounts:
          - name: my-kubernetes-account
            context: gke_1
# ~/.hal/spinnaker-api-team/clouddriver-local.yaml
retrofit:
  loglevel: FULL

redis:
  scheduler: sort

And with a canned clouddriver configuration:

# clouddriver.yaml 
# spinnaker v1.4.0

<<: *clouddriver
# ... lots of details

When Halyard generates this config, it creates the following ~/.spinnaker directory.

.spinnaker/
  clouddriver.yaml
  clouddriver-local.yaml
  orca.yaml
  gate.yaml
  rosco.yaml
  ... # remaining services.yaml

With the contents of clouddriver-local.yaml:

# ~/.spinnaker/clouddriver-local.yaml
retrofit:
  loglevel: FULL

redis:
  scheduler: sort

And the contents of clouddriver.yaml:

# clouddriver.yaml 
# spinnaker v1.4.0

providers: &clouddriver
  kubernetes:
    enabled: true
    accounts:
      - name: my-kubernetes-account
        context: gke_1

<<: *clouddriver
# ... lots of details

Halyard Interface

Halyard will only ever make changes to ~/.hal/config, leaving it up to the user to make optional changes to ~/.hal/<spinnaker-deployment>/*. Below is a list of HTTP methods to be taken against the Halyard daemon by a client.

METHOD PATH /<version>/<deployment>/<type>/<name> BODY DESCRIPTION
POST /accounts account description create new account
POST /webhooks webhook description create new webhook
PUT /enabled boolean enable/disable account
PUT /enabled boolean enable/disable webhook
PUT /accounts/<account> account description edit account
PUT /webhooks/<webhook> account description edit webhook
DELETE /accounts/<account> delete account
DELETE /webhooks/<webhook> delete webhook
GET * return everything matching this path

<version>: Halyard version.

<deployment>: Name of the spinnaker deployment.

<type>: (provider|webhook|git).

<name>: name of provider, webhook, or git entry.

Once an incoming change has been processed, the resulting config will always be run through a validator to ensure that Spinnaker can still be deployed with what's been provided. If the validator passes, the configuration is written out to ~/.spinnaker, as described above. If it fails, the operation will be rejected with an error message describing why.

2. Deploying and Updating Spinnaker

See this document

Non Goals

    1. Replace the Gate API. However, it may be worth having the Halyard CLI provide some very rudimentary operations for starting & examining pipeline state.

Footnotes

[1] 9th factorial divided by the 9th triangle number. Also happens to be an unassigned port close to the range used by other Spinnaker services.