mod auth;
mod cni;
mod config;
mod fiberlb_controller;
mod flashdns_controller;
mod ipam_client;
mod rest;
mod scheduler;
mod services;
mod storage;

use anyhow::Result;
use auth::AuthService;
use chainfire_client::Client as ChainFireClient;
use clap::Parser;
use config::Config;
use ipam_client::IpamClient;
use metrics_exporter_prometheus::PrometheusBuilder;
use k8shost_proto::{
    deployment_service_server::{DeploymentService, DeploymentServiceServer},
    node_service_server::NodeServiceServer,
    pod_service_server::PodServiceServer,
    service_service_server::ServiceServiceServer,
    *,
};
use services::{node::NodeServiceImpl, pod::PodServiceImpl, service::ServiceServiceImpl};
use std::{path::PathBuf, sync::Arc};
use std::time::{SystemTime, UNIX_EPOCH};
use storage::Storage;
use tonic::{transport::Server, Request, Response, Status};
use tracing::{info, warn};
use tracing_subscriber::EnvFilter;

/// k8shost API Server
#[derive(Parser, Debug)]
#[command(name = "k8shost-server")]
#[command(about = "Kubernetes API server for PlasmaCloud's k8shost component")]
struct Args {
    /// Configuration file path
    #[arg(short, long)]
    config: Option<PathBuf>,

    /// Listen address for gRPC server (e.g., "[::]:6443")
    #[arg(long)]
    addr: Option<String>,

    /// Log level (e.g., "info", "debug", "trace")
    #[arg(long)]
    log_level: Option<String>,

    /// FlareDB Placement Driver address (e.g., "127.0.0.1:2479")
    #[arg(long)]
    flaredb_pd_addr: Option<String>,

    /// FlareDB direct address (e.g., "127.0.0.1:50052")
    #[arg(long)]
    flaredb_direct_addr: Option<String>,

    /// ChainFire endpoint for cluster coordination (e.g., "http://127.0.0.1:2379")
    #[arg(long)]
    chainfire_endpoint: Option<String>,

    /// IAM server address (e.g., "http://127.0.0.1:50051")
    #[arg(long)]
    iam_server_addr: Option<String>,

    /// FiberLB server address (e.g., "http://127.0.0.1:50082")
    #[arg(long)]
    fiberlb_server_addr: Option<String>,

    /// FlashDNS server address (e.g., "http://127.0.0.1:50053")
    #[arg(long)]
    flashdns_server_addr: Option<String>,

    /// PrismNET server address (e.g., "http://127.0.0.1:50081")
    #[arg(long)]
    prismnet_server_addr: Option<String>,

    /// Metrics port for Prometheus scraping
    #[arg(long, default_value = "9094")]
    metrics_port: u16,
}

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let args = Args::parse();

    // Load configuration
    let mut settings = ::config::Config::builder()
        .add_source(::config::File::from_str(
            toml::to_string(&Config::default())?.as_str(),
            ::config::FileFormat::Toml,
        ))
        .add_source(::config::Environment::with_prefix("K8SHOST").separator("_"));

    // Add config file if specified
    if let Some(config_path) = &args.config {
        info!("Loading config from file: {}", config_path.display());
        settings = settings.add_source(::config::File::from(config_path.as_path()));
    }

    let loaded_config: Config = settings
        .build()?
        .try_deserialize()
        .map_err(|e| anyhow::anyhow!("Failed to load configuration: {}", e))?;

    // Apply CLI overrides to the loaded configuration
    let config = Config {
        server: config::ServerConfig {
            addr: args
                .addr
                .map(|s| s.parse().unwrap_or(loaded_config.server.addr))
                .unwrap_or(loaded_config.server.addr),
            http_addr: loaded_config.server.http_addr,
            log_level: args.log_level.unwrap_or(loaded_config.server.log_level),
        },
        flaredb: config::FlareDbConfig {
            pd_addr: args.flaredb_pd_addr.or(loaded_config.flaredb.pd_addr),
            direct_addr: args.flaredb_direct_addr.or(loaded_config.flaredb.direct_addr),
        },
        chainfire: config::ChainFireConfig {
            endpoint: args.chainfire_endpoint.or(loaded_config.chainfire.endpoint),
        },
        iam: config::IamConfig {
            server_addr: args.iam_server_addr.unwrap_or(loaded_config.iam.server_addr),
        },
        fiberlb: config::FiberLbConfig {
            server_addr: args.fiberlb_server_addr.unwrap_or(loaded_config.fiberlb.server_addr),
        },
        flashdns: config::FlashDnsConfig {
            server_addr: args.flashdns_server_addr.unwrap_or(loaded_config.flashdns.server_addr),
        },
        prismnet: config::PrismNetConfig {
            server_addr: args
                .prismnet_server_addr
                .unwrap_or(loaded_config.prismnet.server_addr),
        },
    };

    // Initialize tracing
    init_logging(&config.server.log_level);

    info!("Starting k8shost API server on {}", config.server.addr);

    if let Some(endpoint) = &config.chainfire.endpoint {
        let endpoint = endpoint.clone();
        let addr = config.server.addr.to_string();
        tokio::spawn(async move {
            if let Err(error) = register_chainfire_membership(&endpoint, "k8shost", addr).await {
                warn!(error = %error, "ChainFire membership registration failed");
            }
        });
    }

    // Initialize Prometheus metrics exporter
    let metrics_addr = format!("0.0.0.0:{}", args.metrics_port);
    let builder = PrometheusBuilder::new();
    builder
        .with_http_listener(metrics_addr.parse::<std::net::SocketAddr>()?)
        .install()
        .expect("Failed to install Prometheus metrics exporter");

    info!(
        "Prometheus metrics available at http://{}/metrics",
        metrics_addr
    );

    // Initialize FlareDB storage. Prefer direct access when configured, but fall back to
    // the placement-driver path so service startup is resilient to early direct endpoint races.
    let storage = if let Some(addr) = &config.flaredb.direct_addr {
        info!("Connecting to FlareDB directly at {}", addr);
        match Storage::new_direct(addr.clone()).await {
            Ok(s) => {
                info!("Successfully connected to FlareDB (direct)");
                Arc::new(s)
            }
            Err(e) => {
                warn!("Failed direct FlareDB connection: {}", e);
                if let Some(pd_addr) = &config.flaredb.pd_addr {
                    info!("Falling back to FlareDB PD at {}", pd_addr);
                    match Storage::new(pd_addr.clone()).await {
                        Ok(s) => {
                            info!("Successfully connected to FlareDB via PD fallback");
                            Arc::new(s)
                        }
                        Err(pd_error) => {
                            warn!(
                                "Failed to connect to FlareDB via PD fallback: {}. Server will not start.",
                                pd_error
                            );
                            return Err(anyhow::anyhow!(
                                "Failed to connect to FlareDB directly ({}) and via PD fallback ({}).",
                                e,
                                pd_error
                            )
                            .into());
                        }
                    }
                } else {
                    return Err(
                        anyhow::anyhow!("Failed to connect to FlareDB (direct): {}", e).into()
                    );
                }
            }
        }
    } else if let Some(addr) = &config.flaredb.pd_addr {
        info!("Connecting to FlareDB PD at {}", addr);
        match Storage::new(addr.clone()).await {
            Ok(s) => {
                info!("Successfully connected to FlareDB");
                Arc::new(s)
            }
            Err(e) => {
                warn!("Failed to connect to FlareDB: {}. Server will not start.", e);
                return Err(anyhow::anyhow!("Failed to connect to FlareDB: {}", e).into());
            }
        }
    } else {
        return Err(anyhow::anyhow!("No FlareDB address configured.").into());
    };

    // Initialize IAM authentication service
    info!("Connecting to IAM server at {}", config.iam.server_addr);

    let auth_service = match AuthService::new(&config.iam.server_addr).await {
        Ok(s) => {
            info!("Successfully connected to IAM server");
            Arc::new(s)
        }
        Err(e) => {
            warn!("Failed to connect to IAM server: {}. Server will not start.", e);
            return Err(anyhow::anyhow!("Failed to connect to IAM server: {}", e).into());
        }
    };

    // Dedicated runtime for auth interceptors to avoid blocking the main async runtime
    let auth_runtime = Arc::new(tokio::runtime::Runtime::new()?);
    let make_interceptor = |auth: Arc<AuthService>| {
        let rt = auth_runtime.clone();
        move |mut req: Request<()>| -> Result<Request<()>, Status> {
            let auth = auth.clone();
            tokio::task::block_in_place(|| {
                rt.block_on(async move {
                    let tenant_context = auth.authenticate_request(&req).await?;
                    req.extensions_mut().insert(tenant_context);
                    Ok(req)
                })
            })
        }
    };

    // Create IPAM client
    let ipam_client = Arc::new(IpamClient::new(config.prismnet.server_addr.clone()));

    // Create service implementations with storage
    let pod_service = Arc::new(
        PodServiceImpl::new_with_credit_service(storage.clone(), auth_service.clone()).await,
    );
    let service_service = Arc::new(ServiceServiceImpl::new(
        storage.clone(),
        ipam_client,
        auth_service.clone(),
    ));
    let node_service = Arc::new(NodeServiceImpl::new(storage.clone(), auth_service.clone()));
    let deployment_service = DeploymentServiceImpl; // Still unimplemented

    // Start scheduler in background with CreditService integration
    let scheduler = Arc::new(scheduler::Scheduler::new_with_credit_service(storage.clone()).await);
    tokio::spawn(async move {
        scheduler.run().await;
    });
    info!("Scheduler started - tenant-aware with quota enforcement");

    // Start FiberLB controller in background
    let fiberlb_controller = Arc::new(fiberlb_controller::FiberLbController::new(
        storage.clone(),
        config.fiberlb.server_addr.clone(),
        config.iam.server_addr.clone(),
    ));
    tokio::spawn(async move {
        fiberlb_controller.run().await;
    });
    info!("FiberLB controller started - monitoring LoadBalancer services with per-tenant IAM tokens");

    // Start FlashDNS controller in background
    let flashdns_controller = Arc::new(flashdns_controller::FlashDnsController::new(
        storage.clone(),
        config.flashdns.server_addr.clone(),
        config.iam.server_addr.clone(),
    ));
    tokio::spawn(async move {
        flashdns_controller.run().await;
    });
    info!("FlashDNS controller started - managing cluster.local DNS records with per-tenant IAM tokens");

    info!("Starting gRPC server with authentication...");

    // Build gRPC server with authentication layer
    let grpc_server = Server::builder()
        .add_service(
            tonic::codegen::InterceptedService::new(
                PodServiceServer::new(pod_service.as_ref().clone()),
                make_interceptor(auth_service.clone()),
            ),
        )
        .add_service(
            tonic::codegen::InterceptedService::new(
                ServiceServiceServer::new(service_service.as_ref().clone()),
                make_interceptor(auth_service.clone()),
            ),
        )
        .add_service(
            tonic::codegen::InterceptedService::new(
                NodeServiceServer::new(node_service.as_ref().clone()),
                make_interceptor(auth_service.clone()),
            ),
        )
        .add_service(DeploymentServiceServer::new(deployment_service))
        .serve(config.server.addr);

    // HTTP REST API server
    let http_addr = config.server.http_addr;
    let rest_state = rest::RestApiState {
        pod_service: pod_service.clone(),
        service_service: service_service.clone(),
        node_service: node_service.clone(),
        auth_service: auth_service.clone(),
    };
    let rest_app = rest::build_router(rest_state);
    let http_listener = tokio::net::TcpListener::bind(&http_addr).await?;

    info!("k8shost HTTP REST API server starting on {}", http_addr);

    let http_server = async move {
        axum::serve(http_listener, rest_app)
            .await
            .map_err(|e| format!("HTTP server error: {}", e))
    };

    // Run both servers concurrently
    tokio::select! {
        result = grpc_server => {
            result?;
        }
        result = http_server => {
            result?;
        }
    }

    Ok(())
}

// Deployment Service Implementation (placeholder - not part of MVP)
#[derive(Debug, Default)]
struct DeploymentServiceImpl;

#[tonic::async_trait]
impl DeploymentService for DeploymentServiceImpl {
    async fn create_deployment(
        &self,
        _request: Request<CreateDeploymentRequest>,
    ) -> Result<Response<CreateDeploymentResponse>, Status> {
        Err(Status::unimplemented("create_deployment not yet implemented"))
    }

    async fn get_deployment(
        &self,
        _request: Request<GetDeploymentRequest>,
    ) -> Result<Response<GetDeploymentResponse>, Status> {
        Err(Status::unimplemented("get_deployment not yet implemented"))
    }

    async fn list_deployments(
        &self,
        _request: Request<ListDeploymentsRequest>,
    ) -> Result<Response<ListDeploymentsResponse>, Status> {
        Err(Status::unimplemented("list_deployments not yet implemented"))
    }

    async fn update_deployment(
        &self,
        _request: Request<UpdateDeploymentRequest>,
    ) -> Result<Response<UpdateDeploymentResponse>, Status> {
        Err(Status::unimplemented("update_deployment not yet implemented"))
    }

    async fn delete_deployment(
        &self,
        _request: Request<DeleteDeploymentRequest>,
    ) -> Result<Response<DeleteDeploymentResponse>, Status> {
        Err(Status::unimplemented("delete_deployment not yet implemented"))
    }
}

fn init_logging(level: &str) {
    tracing_subscriber::fmt()
        .with_env_filter(EnvFilter::try_from_default_env().unwrap_or_else(|_| EnvFilter::new(level)))
        .init();
}

async fn register_chainfire_membership(
    endpoint: &str,
    service: &str,
    addr: String,
) -> Result<()> {
    let node_id =
        std::env::var("HOSTNAME").unwrap_or_else(|_| format!("{}-{}", service, std::process::id()));
    let ts = SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .unwrap_or_default()
        .as_secs();
    let key = format!("/cluster/{}/members/{}", service, node_id);
    let value = format!(r#"{{"addr":"{}","ts":{}}}"#, addr, ts);
    let deadline = tokio::time::Instant::now() + std::time::Duration::from_secs(120);
    let mut attempt = 0usize;
    let mut last_error = String::new();

    loop {
        attempt += 1;
        match ChainFireClient::connect(endpoint).await {
            Ok(mut client) => match client.put_str(&key, &value).await {
                Ok(_) => return Ok(()),
                Err(error) => last_error = format!("put failed: {}", error),
            },
            Err(error) => last_error = format!("connect failed: {}", error),
        }

        if tokio::time::Instant::now() >= deadline {
            break;
        }

        tracing::warn!(
            attempt,
            endpoint,
            service,
            error = %last_error,
            "retrying ChainFire membership registration"
        );
        tokio::time::sleep(std::time::Duration::from_secs(2)).await;
    }

    anyhow::bail!(
        "failed to register ChainFire membership for {} via {} after {} attempts: {}",
        service,
        endpoint,
        attempt,
        last_error
    )
}