letsbe-sysadmin/app/config.py

"""Agent configuration via environment variables."""

import socket
from functools import lru_cache
from typing import Optional

from pydantic import Field
from pydantic_settings import BaseSettings, SettingsConfigDict

from app import __version__


class Settings(BaseSettings):
    """Agent settings loaded from environment variables.

    All settings are frozen after initialization to prevent runtime mutation.
    """

    model_config = SettingsConfigDict(
        env_file=".env",
        env_file_encoding="utf-8",
        frozen=True,  # Prevent runtime mutation
    )

    # Agent identity
    agent_version: str = Field(default=__version__, description="Agent version for API headers")
    hostname: str = Field(default_factory=socket.gethostname, description="Agent hostname")
    agent_id: Optional[str] = Field(default=None, description="Assigned by orchestrator after registration")

    # Orchestrator connection
    # Default URL is for Docker-based dev where orchestrator runs on the host.
    # When running directly on a Linux tenant server, set ORCHESTRATOR_URL to
    # the orchestrator's public URL (e.g., "https://orchestrator.letsbe.io").
    orchestrator_url: str = Field(
        default="http://host.docker.internal:8000",
        description="Orchestrator API base URL"
    )
    # Token may be None initially; will be set after registration or provided via env
    agent_token: Optional[str] = Field(default=None, description="Authentication token for API calls")

    # Timing intervals (seconds)
    heartbeat_interval: int = Field(default=30, ge=5, le=300, description="Heartbeat interval")
    poll_interval: int = Field(default=5, ge=1, le=60, description="Task polling interval")

    # Logging
    log_level: str = Field(default="INFO", description="Log level (DEBUG, INFO, WARNING, ERROR)")
    log_json: bool = Field(default=True, description="Output logs as JSON")

    # Resilience
    max_concurrent_tasks: int = Field(default=3, ge=1, le=10, description="Max concurrent task executions")
    backoff_base: float = Field(default=1.0, ge=0.1, le=10.0, description="Base backoff time in seconds")
    backoff_max: float = Field(default=60.0, ge=10.0, le=300.0, description="Max backoff time in seconds")
    circuit_breaker_threshold: int = Field(default=5, ge=1, le=20, description="Consecutive failures to trip breaker")
    circuit_breaker_cooldown: int = Field(default=300, ge=30, le=900, description="Cooldown period in seconds")

    # Security - File operations
    allowed_file_root: str = Field(default="/opt/letsbe", description="Root directory for file operations")
    allowed_env_root: str = Field(default="/opt/letsbe/env", description="Root directory for ENV file operations")
    max_file_size: int = Field(default=10 * 1024 * 1024, description="Max file size in bytes (default 10MB)")

    # Security - Shell operations
    shell_timeout: int = Field(default=60, ge=5, le=600, description="Default shell command timeout")

    # Security - Docker operations
    allowed_compose_paths: list[str] = Field(
        default=["/opt/letsbe", "/home/letsbe"],
        description="Allowed directories for compose files"
    )
    allowed_stacks_root: str = Field(
        default="/opt/letsbe/stacks",
        description="Root directory for Docker stack operations"
    )

    # Local persistence
    pending_results_path: str = Field(
        default="~/.letsbe-agent/pending_results.json",
        description="Path for buffering unsent task results"
    )


@lru_cache
def get_settings() -> Settings:
    """Get cached settings instance.

    Settings are loaded once and cached for the lifetime of the process.
    """
    return Settings()