Reference
Console commands
The k8s:* commands the bundle registers: render, dump and make.
Every command the bundle registers, under the k8s: prefix: rendering to standard output, dumping to files, and scaffolding new classes. How to take the output to a cluster is in Rendering and deploying.
The environment for all of them is Symfony's: APP_ENV=prod bin/console k8s:render:apps.
Render: YAML to stdout
| Command | Arguments | Renders |
|---|---|---|
k8s:render:apps | [apps-aliases…] (default: all) | App manifests |
k8s:render:helm | [helm-releases…] (default: all) | Helm releases (adds missing helm repos first) |
k8s:render:all: alias k8s:render | : | everything: Apps first, then releases, one stream |
Output is a multi-document YAML stream, documents joined with --- and each prefixed by a # File: comment naming its virtual path (# File: /shop/api.deployment.yaml). The stream is deterministic and pipe-friendly:
bin/console k8s:render:apps shop | kubectl diff -f -Dump: YAML to files
| Command | Arguments | Dumps |
|---|---|---|
k8s:dump:apps | [apps-aliases…] | App manifests |
k8s:dump:helm | [helm-releases…] | Helm releases |
k8s:dump:all: alias k8s:dump | : | both |
Options (all three):
| Option | Meaning |
|---|---|
-D, --output-dir <dir> | override the configured manifests_dir |
-R, --recreate-output-dir | delete and recreate the target env directory first: guarantees no stale files from removed manifests |
The output base is always suffixed with the current environment, then split by origin:
<manifests_dir>/<env>/apps/<app-alias>/<short-name>.<kind>.yaml
<manifests_dir>/<env>/helm/<release>/<resource-name>.<kind>.yamlKinds are kebab-cased in file names (api.deployment.yaml, session-prune.cron-job.yaml). The # File: headers appear only in rendered streams, not in dumped files: dumped YAML is exactly what kubectl will see.
Without
-R, dumping never deletes: a manifest you removed from code leaves its old file behind. Use-R(or a fresh-Dtarget) for anything that feeds a deploy.
Make: scaffolding
| Command | Arguments | Creates |
|---|---|---|
k8s:make:app <name> | App name without the App suffix (shop → ShopApp) | the App class plus its Manifest/, Container/, and Resources/ directories |
k8s:make:manifest <name> [type] | manifest short name (DNS-subdomain-validated) + kind | a manifest class in the chosen App |
k8s:make:manifest options and behavior:
--app=<alias>selects the target App; omitted, the command asks interactively: as it does for thetypeargument.- Available kinds out of the box:
config-map,cron-job,deployment,ingress,job,secret,service. Kinds without a template (StatefulSet, HPA, PVC, RBAC, Gateway API, …) are written by hand: the manifest types reference is the map. - The generator refuses duplicate short names per kind within an App, and validates the name against Kubernetes' DNS-subdomain rules before writing anything.
Scaffolding targets are controlled by code_dir and namespace_prefix (see the configuration reference); the directory is created by the first command that writes there. The commands write a new class next to the App class they find. In a Phaenon project (a phaenon.json beside composer.json) the App class loads from a compiled output, so the class is written instead as Phaenon, from the templates under templates/phn/, into the source root that output is built from (src/ for .src/), and laid out by the project's own vendor/bin/phn fmt; the command's message names that path. When the formatter fails, the file stays written and the command fails with "<file>" was written, but "phn fmt" could not lay it out: … (what it writes).
The kind list is extensible: manifest generator contexts are tagged services (klass8s.manifest_generator.context), so a platform team can teach k8s:make:manifest its own house kinds: e.g. a worker type scaffolding your AbstractWorkerDeployment subclass.
Exit behavior
Render and dump commands succeed (exit 0) or fail with an uncaught exception: a missing referenced Secret, an invalid name, a failed helm template all abort the render with a stack trace rather than emitting partial output. Make commands print a friendly error and exit 1. There is deliberately no "best effort" mode: partial manifest streams applied to a cluster are worse than no stream. An App with no manifests renders nothing, without an error.
See also
- Rendering & Deploying: workflows built on these commands
- Configuration:
manifests_dir,code_dir, and friends