Permalink
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
150 lines (123 sloc) 6.46 KB
RFC Title Author Status Type Chef-Version
22
Arbitrary Cookbook Identifiers
Daniel DeLeo <ddeleo@getchef.com>
Accepted
Standards Track
12

Arbitrary Cookbook Identifiers

Use Case

The current implementation of cookbook storage in chef server identifies cookbooks with a compound identifier composed of the cookbook's name and version number, where the version number is restricted to the subset of SemVer supported by the dependency resolver subsystem.

This identifier scheme has the following deficiencies:

  • It assumes that users publish content following a similar process to releasing end-user software, when in reality, there are many other workflows. For example, teams that iterate and publish rapidly may prefer to use an auto-incrementing integer scheme, like svn; users that need to publish cookbooks to the server (either manually or as part of a Ci pipeline) to run integration tests may not be ready to assign a static version number to their content; and very basic users (like beginners) may not have a need for versioning at all (yet).

  • Users who have temporarily forked a third party cookbook have no way to publish it to a server without potentially conflicting with the upstream's version.

  • In order to provide a compromise between users who do not need versioning and those who rely on it (via the environments feature, for example), the server allows mutating existing versions of cookbooks, with an option to "freeze" a version. This compromise is often the worst of both worlds, because it requires the user to be diligent about freezing versions and has the potential for catastrophe should the user forget.

Arbitrary Cookbook Identifiers Proposal

This proposal addresses the use case by adding an additional set of APIs for cookbook storage that identify cookbooks by arbitrary identifiers. The endpoint is functionally identical the the current /cookbooks endpoint, with the following exceptions:

  • it is located at BASE_URL/cookbook_artifacts
  • Cookbook "versions" are identified by an arbitrary identifier. This identifier may be limited to 255 URL-safe characters, and there can be other limitations such requiring the first character not be an underscore (so that the server can provide "special" URLs using an underscore in the path component). The URL for a cookbook instance would be BASE_URL/cookbook_artifacts/:cookbook_name/:identifier
  • There is no uniqueness constraint on cookbook name and version; this endpoint allows there to be (e.g.,) multiple distinct Apache2 1.0.0 cookbooks as long as each one has a unique identifier.
  • Cookbooks uploaded to this endpoint are not visible to the /environments/:environment/cookbook_versions endpoint.
  • The endpoint has no concept of "freezing." Overwriting an existing object with the same identifier is always an error.
  • A GET request to /cookbook_artifacts/:cookbook_name should include extra information about each version of a cookbook, such as its SemVer version number, to facilitate tooling that provides a better user experience when working with opaque identifiers. Extended metadata, such as a URL to the upstream source of this cookbook (e.g., supermarket, github, etc.) and associated upstream identifiers (e.g., git commit ID) may also be useful. The entry for an individual cookbook artifact could look like:
  { "name": "apache2",
    "identifier": "886757f9ae3cf2520c82b791195c27ecafd93656",
    "version": "1.10.4",
    "url": "https://chef.example.org/organizations/:org/cookbook_artifacts/apache2/886757f9ae3cf2520c82b791195c27ecafd93656",
    "origin_url": "https://supermarket.getchef.com/cookbooks/apache2/versions/1.10.4/download",
    "origin_id": "1.10.4"
  }
  • If feasible, we should relax validation on the version field of the uploaded cookbook to allow full SemVer version numbers. This allows users to add extra information to the version field if they choose to do so.
  • If feasible, the cookbook_artifacts endpoint should provide a bulk API that allows an API consumer to request multiple cookbook objects in a single request response cycle. The existing server-side dependency solver endpoint at /environments/:environment/cookbook_versions provides this behavior, but is not be used by chef-client when fetching cookbooks via the new artifact-based API.

How Chef Client Uses Cookbooks with Arbitrary IDs

chef-client will have a new mode of operation where it loads a document that contains a list of cookbook artifact names and identifiers to use for a chef-client run. The format of the document may be more complex, but it will contain enough information to produce a list of cookbook name and identifier pairs, e.g.,

{
  "omnibus": "64b3e64306cff223206348e46af545b19032b170",
  "homebrew": "ab4ad2481e08cbb2c4874fd36a44e76f36ec91f7"
}

Given the above, chef-client would make requests to BASE_URL/cookbook_artifacts/omnibus/64b3e64306cff223206348e46af545b19032b170 and BASE_URL/cookbook_artifacts/homebrew/ab4ad2481e08cbb2c4874fd36a44e76f36ec91f7 to get cookbook objects with links to individual files it can download.

This mode is already partially implemented (using currently available APIs) here: https://github.com/chef/chef/blob/9d277e5a4505e5e83e9c4eb30328fdc7148f15c6/lib/chef/policy_builder/policyfile.rb

What Happens to the Existing Cookbooks API

The existing cookbooks API remains as-is, for the following purposes:

  • Backwards compatibility with existing tools and workflows.
  • It provides an "internal supermarket" for users to publish artifacts within their organization.

When both the old and new APIs are used concurrently, cookbooks uploaded to the new end point must, by default, be invisible to the old end point. The new end point allows multiple editions of a cookbook at the same version number to exist simultaneously and provides an implicit guarantee that uploading a cookbook will not interfere with any other active cookbooks.

After a sufficient period of time (one or more major release cycles), we may remove the dependency solver from the chef-server if there is a compelling reason to do so.

Discussion

Any implementation choices in this proposal are open for discussion. The design constraints on the solution are:

  • Must be able to store and fetch cookbooks according to arbitrary identifiers (subject to reasonable constraints on identifier size and URL-safety).
  • Must not impact workflows using the older API, even if both APIs are in use simultaneously.
  • Allowing extended version numbers when using the new API is desirable but not an absolute necessity.