@npm-wharf/fabrik8

fabrik8

Provision and deploy cluster specifications from a single API.

What It's For

fabrik8 was designed to handle initialization of Kubernetes clusters with an initial, known-set of software using a mcgonagall specification.

It works well in environments where you might want ephemeral clusters, clusters on demand (think single tenancy), or think about things like automation and disaster recovery a lot.

What It's Not For

fabrik8 is not a CD solution (at least not presently). It is not meant to be run continuously against the same target (it cannot guarantee 100% idempotence, but makes a best effort to be). Running fabrik8 multiple times may yield unexpected results. For CD solutions, see hikaru.

Approach

fabrik8 uses kubeform, mcgonagall, and hikaru to provision clusters, transform specifications, and deploy them to the newly created cluster.

Environment Variables

As noted in kubeform, many of the environment variables are cloud provider specific and will only be necessary when using a specific provider.

Variable	Description	Default
KUBE_SERVICE	The backing service to use for the request	'GKE'
GOOGLE_APPLICATION_CREDENTIALS	Path to Google API credentials file	''
GOOGLE_ORGANIZATION_ID	Google Organization Id to create projects under	''
GOOGLE_BILLING_ID	Google Billing Account Id to associate with project	''

API

initialize(cluster, specification, data|onData)

The initialize call requires three arguments and returns a promise.

cluster

This will be the same as the kubeform cluster specification (not repeated here).

specification

The specification argument must either be a file path to the spec or a URL to the GitHub repo where the mcgonagall specification is located.

data|onData

The third argument can either be a hash of data required to satisfy tokens present in the specification, or a function that is passed a list of tokens required by the specification. If a function is provided, the result expected is a promise providing a hash of data.

function getTokens (tokenList) {
  // return token hash as a promise
  return Promise.resolve({
    tokenName: tokenValue
  })
}

To control how cluster data will be merged with the mcgonagall specification data, the hash should include a function named onCluster. It will be passed the cluster information returned from kubeform and the data. The signature is:

function onCluster (data, clusterInfo) {
  // assign new properties to data from clusterInfo as needed
  data.someValue = clusterInfo.someSourceValue
}

Without passing this function, all cluster details will be set as children of a .cluster property.

Return

Returns the cluster information from kubeform and the data used to satisfy the specification (under the property specData).

The expectation is that this information will be stored for future retrieval when interacting with the cluster. fabrik8 does not do anything beyond coordinate calls between libraries in order to simplify creation of fully functional

It is recommended that sensitive data (like the Kubernetes admin password) is stored separately in Vault or encrypted before storage.

CLI

A CLI is also provided for fabrik8 that allows you to invoke the API from the command line:

fabrik8 create [--name name] [--url url] --spec ./path/to/spec

Creates a full cluster, reading defaults and existing configuration securely from centralized cluster-info. The only options that are required are configuration for cluster-info, a name or cluster url, and the path to a McGonagall specification. If re-running, only a name is required -- options will be re-read from cluster-info.

• --url, -u the url of the cluster you wish to create, e.g. mycluster.npme.io
• --name, -n the name of the cluster. Can be inferred from the url
• --domain the domain of the cluster. Can be inferred from the url. Defaults to whatever is specified in the cluster-info defaults, if only a name is provided.
• --projectId the name of the gke project to use. Can be inferred from the cluster name
• --environment the environment of the cluster, e.g. development, production
• --specification, -m, --spec the path or URL to the mcgonagall specification
• --verbose output verbose logging (status check output for hikaru)
• --redisUrl the url of the redis containing cluster information. Can also be set through the REDIS_URL environment variable
• --vaultHost the host of the vault server containing sensitive cluster information, auth data, and defaults. Can also be set through the VAULT_HOST environment variable
• --vaultToken an auth token for the vault server. Can also be set through the VAULT_TOKEN environment variable
• --provider the cloud provider to use, defaults to KUBE_SERVICE environment variable or GKE
• --output, -o file to write cluster-info to, for debugging

Values from the defaults can also be overridden as command line args, by prefixing the key with --arg-, e.g. --arg-cluster.worker.memory 26GB, or --arg-common.zones eu-central1-a. Look at the cluster-info defaults for a list of values that can be overridden.

Command line arguments take precedence over saved cluster-info, which take precedence over default cluster-info. Cluster info-will be saved everytime you run fabrik8, so re-running fabrik8 create can be used to change values.

fabrik8 init ./path/to/config -a ./path/to/authFile -s ./path/to/spec -f ./path/to/data -p gke

DEPRECATED

Similar to a blend of CLIs from its component libraries, fabrik8 requires the following arguments:

• ./path/to/config: configuration to base cluster provisioning on
• -a, --auth: the authfile to use when authenticating with the cloud provider
• -m, --spec: the path or URL to the mcgonagall specification
• -f, --data: the path to the data file to supply values for specification tokens
• -o, --output: the file to write data about the new cluster to (defaults to cluster-{date-time}.json)
• -p, --provider: the cloud provider to use
• -s, --scale: a scale factor (if available in the spec)
• -t, --tokens: where to write out all data used to populate the specification
• -v, --apiVersion: the Kubernetes API version (will attempt to detect the default version)
• --verbose: verbose logging, defaults to false

Changelog

1.2.0 (2019-05-04)

Features

• add Vault AppRole support (a5faff0)

<a name="1.1.1"></a>