rust-tokio-supervisor 0.1.4

//! Public configuration input model for supervisor users.
//!
//! The structs in this module are the single raw configuration surface used for
//! YAML loading, template rendering, and JSON Schema generation.

use confique::Config;
use rust_config_tree::ConfigSchema;
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
use std::path::PathBuf;

use crate::{
    config::{
        audit::AuditConfig,
        ipc_security::IpcSecurityConfig,
        policy::{
            ChildStrategyOverrideConfig, DynamicSupervisorConfig, FailureWindowConfig,
            GroupDependencyConfig, GroupStrategyConfig, GroupsConfigSection, MeltdownConfig,
            RestartBudgetConfig, SeverityDefaultConfig, SupervisionPipelineConfig,
        },
    },
    spec::{
        child_declaration::ChildrenConfigSection,
        supervisor::{
            BackpressureConfig, EscalationPolicy, RECOMMENDED_CHANNEL_CAPACITY, SupervisionStrategy,
        },
    },
};

/// Configuration file shape loaded from YAML.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, Config, JsonSchema, ConfigSchema)]
pub struct SupervisorConfig {
    /// Additional configuration files included by `rust-config-tree`.
    #[config(default = [])]
    #[serde(default)]
    pub include: Vec<PathBuf>,

    /// Root supervisor declaration values.
    #[config(nested)]
    pub supervisor: SupervisorRootConfig,

    /// Runtime policy values.
    #[config(nested)]
    #[serde(default)]
    pub policy: PolicyConfig,

    /// Shutdown budget values.
    #[config(nested)]
    #[serde(default)]
    pub shutdown: ShutdownConfig,

    /// Observability switches and capacities.
    #[config(nested)]
    #[serde(default)]
    pub observability: ObservabilityConfig,

    /// Command audit persistence configuration.
    #[config(nested)]
    #[serde(default)]
    pub audit: AuditConfig,

    /// Backpressure policy for observability event subscribers.
    #[config(nested)]
    #[serde(default)]
    pub backpressure: BackpressureConfig,

    /// Group-level restart budgets and group policy declarations.
    #[config(nested)]
    #[serde(default)]
    #[schemars(extend("x-tree-split" = true, "x-tree-transparent-array" = true))]
    pub groups: GroupsConfigSection,

    /// Group-level strategy overrides.
    #[config(default = [])]
    #[serde(default)]
    pub group_strategies: Vec<GroupStrategyConfig>,

    /// Cross-group failure propagation dependencies.
    #[config(default = [])]
    #[serde(default)]
    pub group_dependencies: Vec<GroupDependencyConfig>,

    /// Child-level strategy overrides.
    #[config(default = [])]
    #[serde(default)]
    pub child_strategy_overrides: Vec<ChildStrategyOverrideConfig>,

    /// Default severity class per task role.
    #[config(default = [])]
    #[serde(default)]
    pub severity_defaults: Vec<SeverityDefaultConfig>,

    /// Optional target-side dashboard IPC configuration.
    pub dashboard: Option<DashboardIpcConfig>,

    /// Child declarations loaded from YAML.
    #[config(nested)]
    #[serde(default)]
    #[schemars(extend("x-tree-split" = true, "x-tree-transparent-array" = true))]
    pub children: ChildrenConfigSection,
}

// impl ConfigSchema for SupervisorConfig {
//     /// Returns child configuration paths declared by one loaded layer.
//     ///
//     /// # Arguments
//     ///
//     /// - `layer`: Partially loaded supervisor configuration layer.
//     ///
//     /// # Returns
//     ///
//     /// Returns include paths declared by this configuration layer.
//     fn include_paths(layer: &<Self as Config>::Layer) -> Vec<PathBuf> {
//         layer.include.clone().unwrap_or_default()
//     }
// }

/// Root supervisor configuration.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, Config, JsonSchema)]
pub struct SupervisorRootConfig {
    /// Restart scope strategy for child failures.
    #[config(default = "OneForAll")]
    #[serde(default = "default_supervision_strategy")]
    pub strategy: SupervisionStrategy,
    /// Optional fallback escalation policy used when child and group policies
    /// do not define a more specific policy.
    ///
    /// Root supervisors do not have a parent supervisor at runtime, so
    /// `escalate_to_parent` at root level is a planning and diagnostic label.
    #[schemars(!default)]
    #[serde(default)]
    pub escalation_policy: Option<EscalationPolicy>,
    /// Control command channel capacity.
    ///
    /// Runtime capacity for queued supervisor control commands.
    /// Recommended default: 256.
    #[config(default = 256)]
    #[serde(default = "default_channel_capacity")]
    pub control_channel_capacity: usize,
    /// Event broadcast channel capacity.
    ///
    /// Runtime capacity for supervisor event broadcast delivery.
    /// Recommended default: 256.
    #[config(default = 256)]
    #[serde(default = "default_channel_capacity")]
    pub event_channel_capacity: usize,
    /// Runtime dynamic child acceptance policy.
    #[config(nested)]
    #[serde(default)]
    pub dynamic_supervisor: DynamicSupervisorConfig,
}

/// Restart, backoff, and fuse configuration.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, Config, JsonSchema)]
pub struct PolicyConfig {
    /// Maximum child restarts within the child restart window.
    #[config(default = 10)]
    #[serde(default = "default_child_restart_limit")]
    pub child_restart_limit: u32,
    /// Child restart window in milliseconds.
    #[config(default = 60000)]
    #[serde(default = "default_child_restart_window_ms")]
    pub child_restart_window_ms: u64,
    /// Maximum supervisor failures within the supervisor failure window.
    #[config(default = 30)]
    #[serde(default = "default_supervisor_failure_limit")]
    pub supervisor_failure_limit: u32,
    /// Supervisor failure window in milliseconds.
    #[config(default = 60000)]
    #[serde(default = "default_supervisor_failure_window_ms")]
    pub supervisor_failure_window_ms: u64,
    /// Initial backoff in milliseconds.
    #[config(default = 100)]
    #[serde(default = "default_initial_backoff_ms")]
    pub initial_backoff_ms: u64,
    /// Maximum backoff in milliseconds.
    #[config(default = 5000)]
    #[serde(default = "default_max_backoff_ms")]
    pub max_backoff_ms: u64,
    /// Jitter ratio expressed as a fraction between zero and one.
    #[config(default = 0.10)]
    #[serde(default = "default_jitter_ratio")]
    pub jitter_ratio: f64,
    /// Heartbeat interval in milliseconds.
    #[config(default = 1000)]
    #[serde(default = "default_heartbeat_interval_ms")]
    pub heartbeat_interval_ms: u64,
    /// Stale heartbeat threshold in milliseconds.
    #[config(default = 3000)]
    #[serde(default = "default_stale_after_ms")]
    pub stale_after_ms: u64,
    /// Restart budget used by the supervision pipeline.
    #[config(nested)]
    #[serde(default)]
    pub restart_budget: RestartBudgetConfig,
    /// Failure window used by the supervision pipeline.
    #[config(nested)]
    #[serde(default)]
    pub failure_window: FailureWindowConfig,
    /// Meltdown fuse limits for child, group, and supervisor scopes.
    #[config(nested)]
    #[serde(default)]
    pub meltdown: MeltdownConfig,
    /// Supervision pipeline capacities and concurrent restart limit.
    #[config(nested)]
    #[serde(default)]
    pub supervision_pipeline: SupervisionPipelineConfig,
}

/// Shutdown coordination configuration.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, Config, JsonSchema)]
pub struct ShutdownConfig {
    /// Graceful drain timeout in milliseconds.
    #[config(default = 5000)]
    #[serde(default = "default_graceful_timeout_ms")]
    pub graceful_timeout_ms: u64,
    /// Abort wait timeout in milliseconds.
    #[config(default = 1000)]
    #[serde(default = "default_abort_wait_ms")]
    pub abort_wait_ms: u64,
}

/// Observability configuration.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, Config, JsonSchema)]
pub struct ObservabilityConfig {
    /// Event journal capacity.
    #[config(default = 256)]
    #[serde(default = "default_event_journal_capacity")]
    pub event_journal_capacity: usize,
    /// Whether metrics recording is enabled.
    #[config(default = true)]
    #[serde(default = "default_true")]
    pub metrics_enabled: bool,
    /// Whether command audit recording is enabled.
    #[config(default = true)]
    #[serde(default = "default_true")]
    pub audit_enabled: bool,
}

impl Default for PolicyConfig {
    /// Returns the default runtime policy configuration.
    fn default() -> Self {
        Self {
            child_restart_limit: default_child_restart_limit(),
            child_restart_window_ms: default_child_restart_window_ms(),
            supervisor_failure_limit: default_supervisor_failure_limit(),
            supervisor_failure_window_ms: default_supervisor_failure_window_ms(),
            initial_backoff_ms: default_initial_backoff_ms(),
            max_backoff_ms: default_max_backoff_ms(),
            jitter_ratio: default_jitter_ratio(),
            heartbeat_interval_ms: default_heartbeat_interval_ms(),
            stale_after_ms: default_stale_after_ms(),
            restart_budget: RestartBudgetConfig::default(),
            failure_window: FailureWindowConfig::default(),
            meltdown: MeltdownConfig::default(),
            supervision_pipeline: SupervisionPipelineConfig::default(),
        }
    }
}

impl Default for ShutdownConfig {
    /// Returns the default shutdown coordination configuration.
    fn default() -> Self {
        Self {
            graceful_timeout_ms: default_graceful_timeout_ms(),
            abort_wait_ms: default_abort_wait_ms(),
        }
    }
}

impl Default for ObservabilityConfig {
    /// Returns the default observability configuration.
    fn default() -> Self {
        Self {
            event_journal_capacity: default_event_journal_capacity(),
            metrics_enabled: default_true(),
            audit_enabled: default_true(),
        }
    }
}

/// Returns the default supervision strategy.
fn default_supervision_strategy() -> SupervisionStrategy {
    SupervisionStrategy::OneForAll
}

/// Returns the default child restart limit.
fn default_child_restart_limit() -> u32 {
    10
}

/// Returns the default child restart window in milliseconds.
fn default_child_restart_window_ms() -> u64 {
    60000
}

/// Returns the default supervisor failure limit.
fn default_supervisor_failure_limit() -> u32 {
    30
}

/// Returns the default supervisor failure window in milliseconds.
fn default_supervisor_failure_window_ms() -> u64 {
    60000
}

/// Returns the default initial backoff in milliseconds.
fn default_initial_backoff_ms() -> u64 {
    100
}

/// Returns the default maximum backoff in milliseconds.
fn default_max_backoff_ms() -> u64 {
    5000
}

/// Returns the default jitter ratio.
fn default_jitter_ratio() -> f64 {
    0.10
}

/// Returns the default heartbeat interval in milliseconds.
fn default_heartbeat_interval_ms() -> u64 {
    1000
}

/// Returns the default stale heartbeat threshold in milliseconds.
fn default_stale_after_ms() -> u64 {
    3000
}

/// Returns the default graceful shutdown timeout in milliseconds.
fn default_graceful_timeout_ms() -> u64 {
    5000
}

/// Returns the default abort wait timeout in milliseconds.
fn default_abort_wait_ms() -> u64 {
    1000
}

/// Returns the recommended channel capacity.
fn default_channel_capacity() -> usize {
    RECOMMENDED_CHANNEL_CAPACITY
}

/// Returns the default event journal capacity.
fn default_event_journal_capacity() -> usize {
    256
}

/// Serde default helper: returns true.
fn default_true() -> bool {
    true
}

/// Optional target-side dashboard IPC configuration.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, Config, JsonSchema)]
pub struct DashboardIpcConfig {
    /// Whether the target process opens the local IPC endpoint.
    pub enabled: bool,
    /// Stable target process identifier sent to relay and UI.
    pub target_id: Option<String>,
    /// Local Unix domain socket path used by the target process.
    pub path: Option<PathBuf>,
    /// Socket file permission string such as `0600`.
    pub permissions: Option<String>,
    /// Socket bind behavior when the path already exists.
    pub bind_mode: Option<DashboardIpcBindMode>,
    /// Dynamic registration settings used after IPC is ready.
    pub registration: Option<DashboardRegistrationConfig>,
    /// Optional IPC security pipeline configuration (C1-C9).
    #[schemars(!default)]
    #[serde(default)]
    pub security_config: Option<IpcSecurityConfig>,
}

/// Socket bind behavior for target-side dashboard IPC.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, JsonSchema)]
#[serde(rename_all = "snake_case")]
pub enum DashboardIpcBindMode {
    /// Fail when the socket path already exists.
    CreateNew,
    /// Remove a stale socket path before binding.
    ReplaceStale,
}

/// Dynamic registration settings for a target process.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Config, JsonSchema)]
pub struct DashboardRegistrationConfig {
    /// Whether the target process registers with relay after IPC is ready.
    pub enabled: bool,
    /// Local relay registration socket path.
    pub relay_registration_path: Option<PathBuf>,
    /// Human-readable name shown in the dashboard.
    pub display_name: Option<String>,
    /// Registration lease duration in seconds.
    pub lease_seconds: Option<u64>,
    /// Registration heartbeat interval in seconds.
    pub registration_heartbeat_interval_seconds: Option<u64>,
    /// Timeout in seconds for connecting to the relay registration socket.
    /// Default: 5.
    pub registration_connect_timeout_secs: Option<u64>,
    /// Timeout in seconds for write and ack-read on the registration socket.
    /// Default: 5.
    pub registration_io_timeout_secs: Option<u64>,
}