Component evaluation

How qbec evaluates component code using jsonnet and what it expects the output to look like.

Component loading

  • Consider every .jsonnet, .json, and .yaml file directly under the component directory as a component to be loaded. In this case, the component name is the file name without the extension.
  • Check immediate subdirectories of the component directory to see if they contain an index.jsonnet or index.yaml file. If so, create a component with the sub-directory name.
    • If an index.jsonnet file exists load it for component processing
    • If an index.yaml file exists load all .json and .yaml files in the subdirectory.

Jsonnet evaluation

This works as follows:

  • Collect the list of files to be loaded as described in the previous section.
  • Assuming this leads to 3 files, say, c1.jsonnet, c2.json, and c3.yaml evaluate each file in its own VM in parallel upto a specific concurrency.

The YAML file is parsed as: std.native('parseYaml')(importstr '<file>')

The JSON file is parsed as: std.native('parseJson')(importstr '<file>')

The JSONNET is evaluated in a VM instance as-is. In this case:

  • the extension variable is set to the environment name in question.
  • the extension variable is set to the properties defined for the environment.
  • the extension variable is set to the --app-tag argument passed to the command line (or the empty string, if it wasn’t)
  • the variable is set to the default namespace for the environment. This is typically the namespace defined in the qbec.yaml file for the environment. If the namespaceTagSuffix attribute in qbec.yaml is set to true and an --app-tag argument was specified for the command, the namespace from qbec.yaml is suffixed with the tag with a hyphen in between.
  • all external variables specified from the command line are set.
  • default values for external variables declared in qbec.yaml but not specified on the command line are set.
  • all top level variables associated with the component are set, if specified.

Converting component output to Kubernetes objects

The evaluation above creates a map of component names to outputs returned by the jsonnet, json and yaml files.

The output is allowed to be:

  • A single Kubernetes object (identified as such by virtue of having a kind and apiVersion fields, where kind is not List).
  • A Kubernetes list object where kind is List and it has an items array attribute containing an array of outputs.
  • A map of string keys to outputs
  • An array of outputs

In the latter 3 cases, the output is processed recursively to get to the leaf k8s objects.