plugin-config

name

plugin-config

description

Creates plugin configuration using NemoConfig with environment variables and YAML file support. Use when adding plugin configuration fields, reading config values at runtime, setting up test config overrides, or understanding the env var naming formula. Trigger keywords: config, configuration, NemoConfig, env var, environment variable, plugin_name, NMP_CONFIG, YAML config, config override, test config.

Plugin Configuration (NemoConfig)

Defining Config

Both plugin_name AND plugin_description are required ClassVar[str] fields. Omitting either raises TypeError at class-definition time.

from typing import ClassVar
from pydantic import Field
from nemo_platform_plugin.config import NemoConfig

class MyPluginConfig(NemoConfig):
    plugin_name: ClassVar[str] = "my-plugin"           # kebab-case — matches CLI entry-point key
    plugin_description: ClassVar[str] = "Configuration for my plugin."

    debug: bool = Field(default=False)
    max_workers: int = Field(default=4)
    log_level: str = Field(default="INFO")

Environment Variable Formula

NMP_<SAFE_NAME>_<FIELD> where SAFE_NAME = plugin_name uppercased with non-word chars replaced by _.

plugin_name	Env prefix	Field	Env var
"my-plugin"	NMP_MY_PLUGIN_	debug	NMP_MY_PLUGIN_DEBUG
"agents"	NMP_AGENTS_	runner_backend	NMP_AGENTS_RUNNER_BACKEND
"example"	NMP_EXAMPLE_	greeting_style	NMP_EXAMPLE_GREETING_STYLE

Nested fields use _ as delimiter:

NMP_AGENTS_CONTROLLER_INTERVAL_SECONDS=10
NMP_AGENTS_CONTROLLER_PORT_RANGE_START=49152

YAML Config File

Config key in /etc/nmp/config.yaml equals plugin_name:

my_plugin:
  debug: true
  max_workers: 8

# Helm value path: platformConfig.my_plugin.debug

Override config file location: NMP_CONFIG_FILE_PATH=/my/custom/config.yaml

Priority Order

1. Environment variables (NMP_MY_PLUGIN_*) — highest
2. Config file (/etc/nmp/config.yaml)
3. Pydantic field defaults — lowest

Accessing Config

# Option 1: classmethod (preferred in plugin code)
config = MyPluginConfig.get()

# Option 2: standalone function (preferred when importing one name for all config types)
from nemo_platform_plugin.config import get_nemo_config
config = get_nemo_config(MyPluginConfig)

Both return the same cached singleton. Multiple calls in the same process return the same instance.

Example — ExampleConfig from example-plugin (verbatim)

from typing import ClassVar, Literal
from nemo_platform_plugin.config import NemoConfig
from pydantic import Field

class ExampleConfig(NemoConfig):
    plugin_name: ClassVar[str] = "example"
    plugin_description: ClassVar[str] = "Configuration for the NeMo Platform example plugin."

    greeting_style: Literal["formal", "casual"] = Field(
        default="formal",
        description=(
            "Controls the tone of /hello/{name} responses. "
            '"formal" → "Hello, {name}."  "casual" → "Hey, {name}!"'
            "  Set NMP_EXAMPLE_GREETING_STYLE to override."
        ),
    )
    log_requests: bool = Field(
        default=False,
        description=(
            "When True, each request to the items list endpoint emits a "
            "structured INFO log line including the workspace and page number. "
            "Set NMP_EXAMPLE_LOG_REQUESTS=true to enable."
        ),
    )

Usage across multiple components (no wiring needed):

# In service.py AND controller.py — both call get() independently
config = ExampleConfig.get()

Nested Config Models

Use a plain BaseModel (NOT NemoConfig) for nested sections:

from pydantic import BaseModel, Field
from nemo_platform_plugin.config import NemoConfig

class ControllerConfig(BaseModel):           # plain BaseModel — NOT NemoConfig
    interval_seconds: int = Field(default=5)
    health_check_timeout_seconds: int = Field(default=120)

class AgentsConfig(NemoConfig):
    plugin_name: ClassVar[str] = "agents"
    plugin_description: ClassVar[str] = "Configuration for the NeMo Platform agents plugin."

    controller: ControllerConfig = Field(default_factory=ControllerConfig)
    runner_backend: str = Field(default="in_memory")

Access nested fields: AgentsConfig.get().controller.interval_seconds

Set nested via env: NMP_AGENTS_CONTROLLER_INTERVAL_SECONDS=10

Test Override Pattern

from nemo_platform_plugin.config import (
    set_nemo_config_override,
    clear_nemo_config_override,
    clear_nemo_config_overrides,
)

# In a test:
def test_casual_greeting():
    set_nemo_config_override(ExampleConfig(greeting_style="casual"))
    try:
        config = ExampleConfig.get()
        assert config.greeting_style == "casual"
    finally:
        clear_nemo_config_override(ExampleConfig)

# With autouse pytest fixture (recommended):
import pytest

@pytest.fixture(autouse=True)
def reset_config():
    yield
    clear_nemo_config_overrides()   # clears ALL overrides after each test

Gotchas

• plugin_name must be kebab-case: "my-plugin" not "my_plugin". Use the same value as your CLI entry-point key. The env prefix replaces - with _ automatically.
• Both plugin_name AND plugin_description required: TypeError at class-definition time if either is missing or empty.
• Do NOT call MyPluginConfig() directly: This bypasses env/file loading. Always use MyPluginConfig.get() or get_nemo_config(MyPluginConfig).
• Do NOT set model_config manually: The _NemoConfigMeta metaclass sets it before Pydantic builds the schema. Setting it manually will be overridden and may leave the schema inconsistent.
• Config is cached per process: In tests, always clean up with clear_nemo_config_override(cls) or clear_nemo_config_overrides(). Failing to do so pollutes subsequent tests.
• Abstract intermediates: A class that declares neither plugin_name nor plugin_description is treated as an abstract intermediate (exempt from validation). Declaring plugin_description without plugin_name raises TypeError — you cannot partially fill the requirement.