mentedb-mcp 0.4.21

// Server construction and lifecycle management.
// The MenteDbServer struct and its MCP handler implementation live in tools.rs.
// This module provides the server startup function.

use std::path::Path;

use mentedb::MenteDb;
use rmcp::ServiceExt;
use rmcp::transport::io::stdio;

use crate::config::ServerConfig;
use crate::tools::MenteDbServer;

/// Start the MCP server on stdio transport.
pub async fn run(config: ServerConfig) -> anyhow::Result<()> {
    tracing::info!("Opening MenteDB at {}", config.data_dir.display());

    std::fs::create_dir_all(&config.data_dir)?;

    // Retry opening the database with a timeout to handle lock contention
    // (e.g., another instance is shutting down)
    let db = open_with_retry(
        Path::new(&config.data_dir),
        5,
        std::time::Duration::from_secs(1),
    )?;
    let server = MenteDbServer::new(db, config);

    // Keep a reference to the DB for graceful shutdown
    let db_ref = server.db_ref();
    let db_ref_signal = db_ref.clone();

    // Spawn signal handler for graceful shutdown
    tokio::spawn(async move {
        #[cfg(unix)]
        {
            let mut sigterm =
                tokio::signal::unix::signal(tokio::signal::unix::SignalKind::terminate())
                    .expect("failed to register SIGTERM handler");
            let sigint = tokio::signal::ctrl_c();

            tokio::select! {
                _ = sigterm.recv() => {
                    tracing::info!("Received SIGTERM — flushing database");
                }
                _ = sigint => {
                    tracing::info!("Received SIGINT — flushing database");
                }
            }
        }
        #[cfg(not(unix))]
        {
            tokio::signal::ctrl_c()
                .await
                .expect("failed to register Ctrl+C handler");
            tracing::info!("Received Ctrl+C — flushing database");
        }

        let mut db = db_ref_signal.lock().await;
        if let Err(e) = db.close() {
            tracing::error!(error = %e, "Failed to close database on signal");
        } else {
            tracing::info!("Database flushed on signal");
        }
        std::process::exit(0);
    });

    tracing::info!("Starting MCP server on stdio transport");

    let service = server.serve(stdio()).await.inspect_err(|e| {
        tracing::error!("Server error: {:?}", e);
    })?;

    // Don't propagate error — we must run shutdown even if waiting() fails
    let _ = service.waiting().await;

    // Graceful shutdown: flush indexes, graph, and WAL to disk
    tracing::info!("Shutting down — flushing database to disk");
    let mut db = db_ref.lock().await;
    if let Err(e) = db.close() {
        tracing::error!(error = %e, "Failed to close database cleanly");
    } else {
        tracing::info!("Database closed cleanly");
    }

    tracing::info!("MCP server shut down");
    Ok(())
}

/// Try to open the database, retrying on lock contention.
/// This handles the case where a previous instance is still shutting down.
fn open_with_retry(
    path: &Path,
    max_attempts: u32,
    delay: std::time::Duration,
) -> anyhow::Result<MenteDb> {
    for attempt in 1..=max_attempts {
        match MenteDb::open(path) {
            Ok(db) => return Ok(db),
            Err(e) => {
                let msg = e.to_string();
                if msg.contains("locked") && attempt < max_attempts {
                    tracing::info!(
                        attempt,
                        max_attempts,
                        "Database is locked, waiting for other instance to release (retry in {:?})",
                        delay
                    );
                    std::thread::sleep(delay);
                } else {
                    return Err(anyhow::anyhow!(
                        "Failed to open database at {}: {}",
                        path.display(),
                        msg
                    ));
                }
            }
        }
    }
    unreachable!()
}