prvng_core/nulib/lib_provisioning/services/manager.nu

#!/usr/bin/env nu

# Service Manager Core
# Manages platform service lifecycle, registry, and health checks

use ../config/loader.nu get-config

const SERVICE_STATE_DIR = $"($env.HOME)/.provisioning/services/state"
const SERVICE_PID_DIR = $"($env.HOME)/.provisioning/services/pids"
const SERVICE_LOG_DIR = $"($env.HOME)/.provisioning/services/logs"

# Load service registry from configuration
export def load-service-registry [] -> record {
    let config = (get-config)

    # Load services from config file
    let services_config = $"($config.paths.base)/config/services.toml"

    if not ($services_config | path exists) {
        error make {
            msg: "Service registry not found"
            label: {
                text: $"Missing: ($services_config)"
                span: (metadata $services_config).span
            }
        }
    }

    open $services_config | get services
}

# Get service definition by name
export def get-service-definition [
    service_name: string
] -> record {
    let registry = (load-service-registry)

    if $service_name not-in ($registry | columns) {
        error make {
            msg: $"Service '($service_name)' not found in registry"
            label: {
                text: "Available services: " + (($registry | columns) | str join ", ")
                span: (metadata $service_name).span
            }
        }
    }

    $registry | get $service_name
}

# Check if service is running
export def is-service-running [
    service_name: string
] -> bool {
    let service_def = (get-service-definition $service_name)

    match $service_def.deployment.mode {
        "binary" => {
            let pid_file = $"($SERVICE_PID_DIR)/($service_name).pid"
            if not ($pid_file | path exists) {
                return false
            }

            let pid = (open $pid_file | str trim | into int)
            (ps | where pid == $pid | length) > 0
        }
        "docker" => {
            let container_name = $service_def.deployment.docker.container_name
            let containers = (docker ps --filter $"name=($container_name)" --format "{{.Names}}" | lines)
            $container_name in $containers
        }
        "docker-compose" => {
            let compose_service = $service_def.deployment.docker_compose.service_name
            let project = $service_def.deployment.docker_compose.project_name? | default "provisioning"
            let containers = (docker ps --filter $"label=com.docker.compose.project=($project)" --filter $"label=com.docker.compose.service=($compose_service)" --format "{{.Names}}" | lines)
            ($containers | length) > 0
        }
        "kubernetes" => {
            # Check if deployment exists and has ready pods
            let namespace = $service_def.deployment.kubernetes.namespace
            let deployment = $service_def.deployment.kubernetes.deployment_name
            try {
                let replicas = (kubectl get deployment $deployment -n $namespace -o json | from json | get status.readyReplicas? | default 0)
                $replicas > 0
            } catch {
                false
            }
        }
        "remote" => {
            # For remote services, check health endpoint
            check-service-health $service_name | get healthy
        }
        _ => {
            false
        }
    }
}

# Get service status
export def get-service-status [
    service_name: string
] -> record {
    let is_running = (is-service-running $service_name)
    let service_def = (get-service-definition $service_name)

    let state_file = $"($SERVICE_STATE_DIR)/($service_name).json"
    let state = if ($state_file | path exists) {
        open $state_file
    } else {
        {
            name: $service_name
            status: "unknown"
            health_status: "unknown"
            restart_count: 0
        }
    }

    {
        name: $service_name
        type: $service_def.type
        category: $service_def.category
        status: (if $is_running { "running" } else { "stopped" })
        deployment_mode: $service_def.deployment.mode
        pid: (get-service-pid $service_name)
        uptime: (get-service-uptime $service_name)
        health_status: $state.health_status
        restart_count: $state.restart_count
        auto_start: $service_def.startup.auto_start
        dependencies: $service_def.dependencies
    }
}

# Get service PID
def get-service-pid [
    service_name: string
] -> int {
    let pid_file = $"($SERVICE_PID_DIR)/($service_name).pid"

    if not ($pid_file | path exists) {
        return 0
    }

    try {
        open $pid_file | str trim | into int
    } catch {
        0
    }
}

# Get service uptime in seconds
def get-service-uptime [
    service_name: string
] -> int {
    let state_file = $"($SERVICE_STATE_DIR)/($service_name).json"

    if not ($state_file | path exists) {
        return 0
    }

    let state = (open $state_file)

    if "started_at" not-in $state {
        return 0
    }

    try {
        let started = ($state.started_at | into datetime)
        let now = (date now)
        (($now - $started) | into int) / 1_000_000_000  # Convert nanoseconds to seconds
    } catch {
        0
    }
}

# Start service
export def start-service [
    service_name: string
    --force (-f)
] -> bool {
    # Ensure state directories exist
    mkdir $SERVICE_STATE_DIR
    mkdir $SERVICE_PID_DIR
    mkdir $SERVICE_LOG_DIR

    if (is-service-running $service_name) and not $force {
        print $"Service '($service_name)' is already running"
        return true
    }

    if $force and (is-service-running $service_name) {
        print $"Stopping existing instance of '($service_name)'..."
        stop-service $service_name --force
        sleep 2sec
    }

    let service_def = (get-service-definition $service_name)

    # Check dependencies
    for dep in $service_def.dependencies {
        if not (is-service-running $dep) {
            print $"Starting dependency: ($dep)"
            start-service $dep
        }
    }

    print $"Starting service: ($service_name)"

    # Delegate to lifecycle module
    use lifecycle.nu start-service-by-mode
    let result = (start-service-by-mode $service_def $service_name)

    if $result {
        # Update state
        let state = {
            name: $service_name
            status: "running"
            started_at: (date now | format date "%Y-%m-%dT%H:%M:%S%z")
            health_status: "unknown"
            restart_count: 0
        }

        $state | to json | save -f $"($SERVICE_STATE_DIR)/($service_name).json"

        # Wait for health check
        print $"Waiting for ($service_name) to become healthy..."
        use health.nu wait-for-service
        wait-for-service $service_name $service_def.startup.start_timeout
    } else {
        print $"Failed to start service: ($service_name)"
        false
    }
}

# Stop service
export def stop-service [
    service_name: string
    --force (-f)
] -> bool {
    if not (is-service-running $service_name) {
        print $"Service '($service_name)' is not running"
        return true
    }

    print $"Stopping service: ($service_name)"

    let service_def = (get-service-definition $service_name)

    # Delegate to lifecycle module
    use lifecycle.nu stop-service-by-mode
    let result = (stop-service-by-mode $service_name $service_def $force)

    if $result {
        # Update state
        let state_file = $"($SERVICE_STATE_DIR)/($service_name).json"
        if ($state_file | path exists) {
            let state = (open $state_file)
            {
                ...$state
                status: "stopped"
                stopped_at: (date now | format date "%Y-%m-%dT%H:%M:%S%z")
            } | to json | save -f $state_file
        }

        # Remove PID file
        let pid_file = $"($SERVICE_PID_DIR)/($service_name).pid"
        if ($pid_file | path exists) {
            rm $pid_file
        }
    }

    $result
}

# Restart service
export def restart-service [
    service_name: string
] -> bool {
    print $"Restarting service: ($service_name)"

    if (is-service-running $service_name) {
        stop-service $service_name --force
        sleep 2sec
    }

    start-service $service_name
}

# Check service health
export def check-service-health [
    service_name: string
] -> record {
    let service_def = (get-service-definition $service_name)

    use health.nu perform-health-check
    perform-health-check $service_name $service_def.health_check
}

# Wait for service to be healthy
export def wait-for-service [
    service_name: string
    timeout: int = 60
] -> bool {
    use health.nu wait-for-service
    wait-for-service $service_name $timeout
}

# Get all services
export def list-all-services [] -> list {
    let registry = (load-service-registry)
    $registry | columns | each { |name|
        get-service-status $name
    }
}

# Get running services
export def list-running-services [] -> list {
    list-all-services | where status == "running"
}

# Get service logs
export def get-service-logs [
    service_name: string
    --lines: int = 50
    --follow (-f)
] -> string {
    let log_file = $"($SERVICE_LOG_DIR)/($service_name).log"

    if not ($log_file | path exists) {
        return $"No logs found for ($service_name)"
    }

    if $follow {
        tail -f -n $lines $log_file
    } else {
        tail -n $lines $log_file
    }
}

# Initialize service state directories
export def init-service-state [] {
    mkdir $SERVICE_STATE_DIR
    mkdir $SERVICE_PID_DIR
    mkdir $SERVICE_LOG_DIR
}