custom-charm

name	custom-charm
description	End-to-end workflow for building ops-framework charms for custom applications (K8s and machine)

Custom Application Charm Workflow

This skill covers building a Juju charm for an application that does not fit a recognised 12-factor PaaS framework (Flask, Django, FastAPI, Go, Express, Spring Boot). These charms use the full ops framework — you write src/charm.py yourself, managing the workload lifecycle directly via Pebble (K8s) or systemd/apt (machine).

When to Use This Path

Use this path (Path B) when:

• The application is not one of the six PaaS-supported frameworks
• You need full control over the workload lifecycle (custom start sequences, multi-process, etc.)
• The workload requires non-standard packaging (not a simple WSGI/ASGI/HTTP server)
• The user explicitly requests a custom ops-framework charm

Do not use this path for:

• Flask, Django, FastAPI, Go, Express, Spring Boot → use the twelve-factor skill instead
• Infrastructure software (databases, caches, message brokers) → Path C

Substrate Decision Matrix

Choose the substrate before scaffolding. Ask the user to confirm.

Signal	Substrate	Reasoning
Dockerfile or OCI image exists	K8s	Already containerised; natural fit for Pebble
Cloud-native / stateless workload	K8s	Designed for orchestration
Bare metal / GPU / kernel modules	Machine	Needs direct hardware access
systemd service files present	Machine	Already designed for systemd
Heavy OS-level integration (PAM, cron, udev)	Machine	Relies on host OS services
No strong signal	K8s (default)	Easier to distribute and test

Step-by-Step Workflow

1. Analyse the Application

Run analyse_framework on the source directory. When no PaaS framework is detected, the tool returns workload_hints indicating whether a Dockerfile, docker-compose file, or systemd service files are present, along with a suggested_substrate.

2. Confirm the Substrate

Present the substrate recommendation to the user with reasoning. Wait for confirmation before proceeding — this choice affects everything downstream.

3. Scaffold the Charm

Use charmcraft_init with the appropriate profile:

• K8s: profile: "kubernetes"
• Machine: profile: "machine"

charmcraft init --profile=kubernetes --name=my-app

This creates the base structure: charmcraft.yaml, src/charm.py, tox.ini, and test directories.

4. Customise charmcraft.yaml

K8s Charm

Add containers, resources, and storage:

name: my-app
type: charm
bases:
  - build-on:
      - name: ubuntu
        channel: "22.04"
    run-on:
      - name: ubuntu
        channel: "22.04"

containers:
  my-app:
    resource: oci-image

resources:
  oci-image:
    type: oci-image
    description: OCI image for the application

storage:
  data:
    type: filesystem
    minimum-size: 1G
    location: /data

requires:
  tracing:
    interface: tracing
    limit: 1

Machine Charm

Machine charms typically need no containers or OCI resources:

name: my-app
type: charm
bases:
  - build-on:
      - name: ubuntu
        channel: "22.04"
    run-on:
      - name: ubuntu
        channel: "22.04"

storage:
  data:
    type: filesystem
    minimum-size: 1G
    location: /data

requires:
  tracing:
    interface: tracing
    limit: 1

5. Write src/charm.py

K8s Charm (Pebble)

#!/usr/bin/env python3
"""Charm for my-app on Kubernetes."""

import ops
import ops_tracing


class MyAppCharm(ops.CharmBase):
    """Charm the application using Pebble."""

    def __init__(self, framework: ops.Framework):
        super().__init__(framework)
        self._tracing = ops_tracing.Tracing(self, "tracing")
        self._container = self.unit.get_container("my-app")

        framework.observe(self.on["my-app"].pebble_ready, self._on_pebble_ready)
        framework.observe(self.on.config_changed, self._on_config_changed)

    def _on_pebble_ready(self, event: ops.PebbleReadyEvent) -> None:
        """Start the workload when the container is ready."""
        self._container.add_layer("my-app", self._pebble_layer(), combine=True)
        self._container.autostart()
        self.unit.status = ops.ActiveStatus()

    def _on_config_changed(self, event: ops.ConfigChangedEvent) -> None:
        """Update the workload config and restart if running."""
        if not self._container.can_connect():
            event.defer()
            return
        self._container.add_layer("my-app", self._pebble_layer(), combine=True)
        self._container.replan()
        self.unit.status = ops.ActiveStatus()

    def _pebble_layer(self) -> ops.pebble.LayerDict:
        """Return the Pebble layer configuration."""
        return {
            "summary": "my-app layer",
            "description": "Pebble layer for my-app",
            "services": {
                "my-app": {
                    "override": "replace",
                    "summary": "my-app service",
                    "command": "/usr/bin/my-app",
                    "startup": "enabled",
                    "environment": {},
                },
            },
        }


if __name__ == "__main__":
    ops.main(MyAppCharm)

Key patterns for K8s charms:

• Get the container with self.unit.get_container("container-name")
• Observe pebble-ready — this fires when the container's Pebble agent becomes reachable
• Use add_layer() + autostart() on first start; add_layer() + replan() on changes
• Check self._container.can_connect() before interacting with Pebble; defer if not ready
• Use self._container.push() to write config files into the container
• Use self._container.exec() to run commands inside the container

Machine Charm (systemd)

#!/usr/bin/env python3
"""Charm for my-app on machines."""

import subprocess

import ops
import ops_tracing


class MyAppCharm(ops.CharmBase):
    """Charm the application using apt and systemd."""

    def __init__(self, framework: ops.Framework):
        super().__init__(framework)
        self._tracing = ops_tracing.Tracing(self, "tracing")

        framework.observe(self.on.install, self._on_install)
        framework.observe(self.on.start, self._on_start)
        framework.observe(self.on.config_changed, self._on_config_changed)

    def _on_install(self, event: ops.InstallEvent) -> None:
        """Install packages and prepare the system."""
        self.unit.status = ops.MaintenanceStatus("Installing dependencies")
        subprocess.check_call(["apt-get", "update", "-y"])
        subprocess.check_call(["apt-get", "install", "-y", "my-app"])

    def _on_start(self, event: ops.StartEvent) -> None:
        """Start the workload service."""
        subprocess.check_call(["systemctl", "enable", "--now", "my-app"])
        self.unit.status = ops.ActiveStatus()

    def _on_config_changed(self, event: ops.ConfigChangedEvent) -> None:
        """Update configuration and restart the service."""
        self._write_config()
        subprocess.check_call(["systemctl", "restart", "my-app"])
        self.unit.status = ops.ActiveStatus()

    def _write_config(self) -> None:
        """Write the application config file from charm config."""
        # Read charm config and write to the appropriate location.
        # Example: /etc/my-app/config.yaml
        pass


if __name__ == "__main__":
    ops.main(MyAppCharm)

Key patterns for machine charms:

• Use install event for package installation (apt-get, snap install)
• Use start event to enable and start the systemd service
• Write config files to standard locations (/etc/my-app/, etc.)
• Use subprocess.check_call() for system commands
• Use systemctl restart after config changes

6. Common Patterns

Status Management

Set status to communicate the charm's state to the operator:

# During long operations.
self.unit.status = ops.MaintenanceStatus("Installing packages")

# When waiting for a relation.
self.unit.status = ops.WaitingStatus("Waiting for database relation")

# When a required config option is missing.
self.unit.status = ops.BlockedStatus("Missing required config: database-url")

# When everything is working.
self.unit.status = ops.ActiveStatus()

Config Validation

Validate config early and set BlockedStatus on invalid input:

def _on_config_changed(self, event: ops.ConfigChangedEvent) -> None:
    port = self.config.get("port")
    if not isinstance(port, int) or not (1 <= port <= 65535):
        self.unit.status = ops.BlockedStatus("Invalid port number")
        return
    # Apply the valid config...

Relation Handling

Read and write relation data in event handlers:

def _on_database_relation_changed(self, event: ops.RelationChangedEvent) -> None:
    if not event.unit:
        return
    host = event.relation.data[event.unit].get("host")
    port = event.relation.data[event.unit].get("port")
    if not host or not port:
        self.unit.status = ops.WaitingStatus("Waiting for database credentials")
        return
    # Configure the workload with the database connection...

For detailed relation data patterns, load the relation-data-design skill. For OAuth / OIDC / login flows backed by the Canonical Identity Platform, load the identity-platform skill — it covers the oauth, oauth-cli, oidc-info, and hydra-token-introspect relations and how the secret-backed client credentials flow through to the workload.

Storage

Access Juju-managed storage in event handlers:

def _on_storage_attached(self, event: ops.StorageAttachedEvent) -> None:
    storage_path = event.storage.location
    # Use storage_path for persistent data...

Secrets (Juju 3.x)

Use Juju secrets for sensitive data:

def _on_secret_changed(self, event: ops.SecretChangedEvent) -> None:
    secret = event.secret
    content = secret.get_content(refresh=True)
    password = content.get("password")
    # Apply the new secret...

7. Build and Deploy

Custom charms do not need rockcraft. The cycle is:

charmcraft pack
juju deploy ./my-app_amd64.charm

For K8s charms with an OCI image resource:

charmcraft pack
juju deploy ./my-app_amd64.charm --resource oci-image=docker.io/library/my-app:latest

Use the charmcraft_pack and juju_deploy tools. For K8s charms, pass the OCI image via resources={"oci-image": "..."}.

8. OCI Image Selection (K8s Charms)

K8s charms need an OCI image for the container resource. There are two options:

Option A: Use an existing registry image

1. Search Docker Hub: registry_search(query="<workload>")
2. Evaluate candidates: registry_image_info(image="<name>", tag="<version>")
3. Prefer official images, recent updates, specific version tags, and amd64+arm64 support
4. Confirm the image choice with the user
5. Deploy with the image reference: juju_deploy(..., resources={"oci-image": "docker.io/library/<name>:<tag>"})

Option B: Build a rock

When no suitable image exists or the workload needs custom packaging:

1. Write rockcraft.yaml for the application
2. rockcraft_pack to build the rock
3. skopeo_registry_push to push to the local registry
4. Deploy with the local image reference

Option A is the common case for custom charms wrapping well-known software. Option B is needed when the application is custom code or requires modifications to the base image.

9. Fast Dev Cycle

For Python source changes only, use the fast path:

1. Edit src/charm.py (or files in src/ and lib/)
2. charm_sync to push changes to the running unit
3. juju_dispatch to fire an event (e.g. config-changed, update-status)

Use the full charmcraft_pack → juju_refresh cycle when:

• charmcraft.yaml changed (new relations, config options, resources)
• Dependencies changed (requirements.txt)
• Before declaring the charm done

10. Generate Tests

Unit tests: Load the scenario-tests skill. Cover:

• install / start → ActiveStatus
• pebble-ready (K8s) → Pebble layer applied, service started
• config-changed → config applied, service restarted
• Relation events → data read/written correctly
• Actions (if any)

Integration tests: Load the jubilant-tests skill. Cover:

• Deploy → ActiveStatus
• Relate to required integrations
• Config changes take effect
• Actions produce expected results

11. Validate

Run charm_validate before declaring the charm complete. This runs unit tests and charmcraft pack, producing a pass/fail checklist.

Re-Deploying After Changes

charmcraft_pack → juju_refresh --path → juju_wait → verify

No rockcraft or registry push needed for custom charms (unlike 12-factor).

Common Pitfalls

1. Pebble layer hygiene — add_layer() without combine=True, Pebble methods called without a can_connect() guard, or service entries missing override / command / startup are all checked deterministically by charmlint (PEB001 / PEB002 / PEB003). Run charmlint to surface them — the skill body no longer recites the rules.

2. Blocking the hook with long-running commands — hooks have a timeout (default 5 minutes). For long installations, consider breaking work across events or increasing the timeout.

3. Writing config directly in the install hook — config values may not be set yet during install. Use config-changed for applying configuration.

4. Using kubernetes profile for a PaaS framework — if the app is Flask/Django/etc., use the framework profile with the twelve-factor skill instead. The kubernetes profile gives a bare ops charm.

5. Missing OCI image resource for K8s charms — the containers section in charmcraft.yaml must reference a resource, and that resource must be declared in the resources section.

Troubleshooting

Symptom	Likely Cause	Fix
Unit stuck in waiting with "waiting for Pebble"	Container not started or image pull failed	Check juju debug-log; verify OCI image resource is correct
pebble-ready never fires	Container name mismatch between charmcraft.yaml and charm code	Ensure the container name in self.on["name"] matches charmcraft.yaml
ConnectionError from Pebble	Calling Pebble methods before can_connect() returns True	Add a can_connect() guard and defer the event
Service crashes after replan()	Bad command or missing binary in the container	Check the Pebble layer command; use container.exec() to debug
apt-get fails during install (machine)	Missing universe repo or network issue	Add add-apt-repository or check proxy settings
systemctl fails (machine)	Service file not installed or unit not enabled	Verify the package installs a .service file; use systemctl status
Charm goes to error on relation event	Accessing relation data that does not exist yet	Guard with if not event.unit: return and check for required keys
Config change has no effect	Forgot to restart the service after writing new config	Call replan() (K8s) or systemctl restart (machine) after applying config