sanctum/tests/live_crash_resilience_test.rs

use std::path::PathBuf;
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Arc;

use bytes::Bytes;
use dav_server::{
    davpath::DavPath,
    fs::{DavFileSystem, OpenOptions},
};
use rand::RngCore;
use sanctum::{
    crypto::{derive_kek, generate_dek, generate_salt, wrap_dek, KdfParams, FORMAT_VERSION},
    storage::Database,
    verify::verify_container,
    vfs::SanctumFs,
};

/// Live-Crash- und Stresstest: Simuliert harten Verbindungsabbruch und Power-Cut
/// während intensiver paralleler Schreibvorgänge im VFS.
#[tokio::test]
async fn test_live_crash_and_recovery_stress() {
    let temp_dir = std::env::temp_dir();
    let container_path: PathBuf = temp_dir.join(format!(
        "sanctum_live_stress_{}.sanctum",
        std::process::id()
    ));
    if container_path.exists() {
        let _ = std::fs::remove_file(&container_path);
    }

    let password = "LiveStressPassword2026!";
    let salt = generate_salt();
    let kdf_params = KdfParams {
        memory_cost: sanctum::crypto::MIN_MEMORY_COST_KIB,
        time_cost: sanctum::crypto::MIN_TIME_COST,
        parallelism: 1,
    };
    let kek = derive_kek(password, &salt, &kdf_params).expect("KEK derivation");
    let dek = generate_dek();
    let (wrapped_dek, header_nonce, header_tag) = wrap_dek(&kek, &dek).expect("DEK wrapping");

    // 1. Initialisierung des Containers
    {
        let db = Database::open(&container_path).expect("Open database");
        db.init_schema(&salt, &kdf_params, &wrapped_dek, &header_nonce, &header_tag)
            .expect("Init schema");
        db.checkpoint().expect("Initial Checkpoint");
    }

    // 2. Parallele Schreiblast mit SanctumFs erzeugen
    let stop_signal = Arc::new(AtomicBool::new(false));
    let db = Database::open(&container_path).expect("Open database for VFS");
    let fs = SanctumFs::new(db, dek.clone(), FORMAT_VERSION);

    let mut handles = Vec::new();

    // Spawn 4 parallele Schreiber
    for worker_id in 0..4 {
        let fs_clone = fs.clone();
        let stop_clone = stop_signal.clone();

        let handle = tokio::spawn(async move {
            let mut file_idx = 0;
            while !stop_clone.load(Ordering::Relaxed) && file_idx < 10 {
                let file_path_str = format!("/worker_{}_file_{}.dat", worker_id, file_idx);
                let dav_path = DavPath::new(&file_path_str).unwrap();

                let mut opts = OpenOptions::default();
                opts.write = true;
                opts.create = true;
                opts.truncate = true;

                // Datei anlegen
                let mut file = match fs_clone.open(&dav_path, opts).await {
                    Ok(f) => f,
                    Err(_) => break,
                };

                // Mehrere 256-KB Blöcke schreiben (über mehrere Chunks hinweg)
                let mut payload = vec![0u8; 256 * 1024];
                rand::thread_rng().fill_bytes(&mut payload);

                for _ in 0..6 {
                    if stop_clone.load(Ordering::Relaxed) {
                        break;
                    }
                    let _ = file.write_bytes(Bytes::copy_from_slice(&payload)).await;
                }
                let _ = file.flush().await;
                file_idx += 1;
            }
        });
        handles.push(handle);
    }

    // Lass die Worker 500ms unter Volllast schreiben
    tokio::time::sleep(tokio::time::Duration::from_millis(500)).await;

    // 3. Simuliere abrupten Prozessabbruch (Hard Kill / Power Cut)
    // Wir brechen die Worker hart ab (Cancel) und verwerfen das FS-Handle ohne sauberen Unmount
    stop_signal.store(true, Ordering::SeqCst);
    for h in handles {
        h.abort(); // Simuliert Kill
    }

    // FS ohne Checkpoint/Drop-Finalisierung freigeben
    drop(fs);

    // 4. Recovery & Integritätsprüfung nach Crash
    // Das System muss die SQLite WAL-Datei automatisch erkennen und verarbeiten
    let verify_result =
        verify_container(&container_path, Some(&dek), false).expect("Verify post-crash");
    assert!(
        verify_result.is_healthy(),
        "Container muss nach Crash vollkommen konsistent sein! Fehler: {:?}",
        verify_result.errors
    );
    assert_eq!(
        verify_result.corrupted_chunks, 0,
        "Keine korrupten Chunks erlaubt"
    );

    // 5. Konsistentes Weiterarbeiten nach dem Absturz
    let db_recovered = Database::open(&container_path).expect("Open database after crash");
    let fs_recovered = SanctumFs::new(db_recovered, dek.clone(), FORMAT_VERSION);

    // Neue Datei im wiederhergestellten Dateisystem anlegen und lesen
    let recovery_test_path = DavPath::new("/post_crash_verification.txt").unwrap();
    {
        let mut opts = OpenOptions::default();
        opts.write = true;
        opts.create = true;
        opts.truncate = true;
        let mut file = fs_recovered
            .open(&recovery_test_path, opts)
            .await
            .expect("Create post-crash file");
        file.write_bytes(Bytes::from_static(b"Sanctum Crash Consistency Verified!"))
            .await
            .expect("Write post crash file");
        file.flush().await.expect("Flush post crash file");
    }

    // Datei wieder einlesen
    {
        let mut opts = OpenOptions::default();
        opts.read = true;
        let mut file = fs_recovered
            .open(&recovery_test_path, opts)
            .await
            .expect("Read post-crash file");
        let bytes = file.read_bytes(1024).await.expect("Read bytes");
        assert_eq!(&bytes[..], b"Sanctum Crash Consistency Verified!");
    }

    // Sauberes Aufräumen der Testdatei
    drop(fs_recovered);
    let _ = std::fs::remove_file(&container_path);
}