synops/maskinrommet/src/maintenance.rs

// Graceful shutdown — vedlikeholdsmodus med koordinert nedstengning.
//
// Flyt:
// 1. Admin kaller initiate_maintenance med tidspunkt
// 2. System oppretter systemvarsel → frontend viser nedtelling via PG NOTIFY
// 3. Bakgrunnsoppgave venter til vedlikeholdstidspunkt
// 4. Setter maintenance_active → blokkerer nye LiveKit-rom + jobbkø stopper dequeue
// 5. Venter på at kjørende jobber fullføres (med timeout)
// 6. Avslutter prosessen → systemd restarter
//
// Ref: docs/concepts/adminpanelet.md, oppgave 15.2

use chrono::{DateTime, Utc};
use serde::Serialize;
use sqlx::PgPool;
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Arc;
use tokio::sync::Mutex;
use uuid::Uuid;

/// Delt vedlikeholdstilstand — klones inn i AppState.
#[derive(Clone)]
pub struct MaintenanceState {
    /// Satt til true når vedlikeholdstidspunktet er nådd.
    /// Når true: jobbkø slutter å dequeue, nye LiveKit-tokens avvises.
    pub active: Arc<AtomicBool>,

    /// Satt til true når admin har initiert vedlikehold (men tidspunktet
    /// trenger ikke være nådd ennå). Brukes for å vise status.
    pub initiated: Arc<AtomicBool>,

    /// Vedlikeholdstidspunkt og varsel-node-id.
    inner: Arc<Mutex<MaintenanceInner>>,
}

struct MaintenanceInner {
    scheduled_at: Option<DateTime<Utc>>,
    announcement_node_id: Option<Uuid>,
    initiated_by: Option<Uuid>,
    /// Handle for å avbryte den planlagte shutdown-tasken.
    abort_handle: Option<tokio::task::AbortHandle>,
}

/// Status-respons for admin-panelet.
#[derive(Serialize)]
pub struct MaintenanceStatus {
    pub initiated: bool,
    pub active: bool,
    pub scheduled_at: Option<String>,
    pub announcement_node_id: Option<Uuid>,
    pub initiated_by: Option<Uuid>,
    pub running_jobs: Vec<RunningJob>,
}

#[derive(Serialize)]
pub struct RunningJob {
    pub id: Uuid,
    pub job_type: String,
    pub started_at: Option<String>,
    pub collection_node_id: Option<Uuid>,
}

impl MaintenanceState {
    pub fn new() -> Self {
        Self {
            active: Arc::new(AtomicBool::new(false)),
            initiated: Arc::new(AtomicBool::new(false)),
            inner: Arc::new(Mutex::new(MaintenanceInner {
                scheduled_at: None,
                announcement_node_id: None,
                initiated_by: None,
                abort_handle: None,
            })),
        }
    }

    /// Er vedlikeholdsmodus aktivert? (Tidspunktet er nådd.)
    pub fn is_active(&self) -> bool {
        self.active.load(Ordering::Relaxed)
    }

    /// Er vedlikehold initiert? (Planlagt, men kanskje ikke nådd ennå.)
    pub fn is_initiated(&self) -> bool {
        self.initiated.load(Ordering::Relaxed)
    }

    /// Hent full status inkludert kjørende jobber.
    pub async fn status(&self, db: &PgPool) -> Result<MaintenanceStatus, sqlx::Error> {
        let inner = self.inner.lock().await;
        let running_jobs = fetch_running_jobs(db).await?;

        Ok(MaintenanceStatus {
            initiated: self.is_initiated(),
            active: self.is_active(),
            scheduled_at: inner.scheduled_at.map(|dt| dt.to_rfc3339()),
            announcement_node_id: inner.announcement_node_id,
            initiated_by: inner.initiated_by,
            running_jobs,
        })
    }

    /// Initier vedlikehold: sett tidspunkt, opprett varsel, start nedtelling.
    pub async fn initiate(
        &self,
        db: &PgPool,
        scheduled_at: DateTime<Utc>,
        initiated_by: Uuid,
    ) -> Result<Uuid, String> {
        if self.is_initiated() {
            return Err("Vedlikehold er allerede initiert".to_string());
        }

        // Opprett systemvarsel
        let node_id = Uuid::now_v7();

        let metadata = serde_json::json!({
            "announcement_type": "critical",
            "scheduled_at": scheduled_at.to_rfc3339(),
            "blocks_new_sessions": true,
            "maintenance_shutdown": true,
        });

        // PG — persistent lagring (NOTIFY-trigger sender sanntidsoppdatering)
        sqlx::query(
            r#"INSERT INTO nodes (id, node_kind, title, content, visibility, metadata, created_by)
               VALUES ($1, 'system_announcement', 'Planlagt vedlikehold',
                       'Systemet stenges for vedlikehold. Lagre arbeidet ditt.',
                       'open', $2, $3)"#,
        )
        .bind(node_id)
        .bind(&metadata)
        .bind(initiated_by)
        .execute(db)
        .await
        .map_err(|e| format!("PG-feil: {e}"))?;

        tracing::info!(
            announcement_id = %node_id,
            scheduled_at = %scheduled_at,
            initiated_by = %initiated_by,
            "Vedlikehold initiert"
        );

        // Start bakgrunnsoppgave for shutdown-koordinering
        let state = self.clone();
        let db2 = db.clone();
        let handle = tokio::spawn(async move {
            shutdown_coordinator(state, db2, scheduled_at, node_id).await;
        });

        // Lagre tilstand
        let mut inner = self.inner.lock().await;
        inner.scheduled_at = Some(scheduled_at);
        inner.announcement_node_id = Some(node_id);
        inner.initiated_by = Some(initiated_by);
        inner.abort_handle = Some(handle.abort_handle());
        self.initiated.store(true, Ordering::Relaxed);

        Ok(node_id)
    }

    /// Avbryt planlagt vedlikehold.
    pub async fn cancel(
        &self,
        db: &PgPool,
    ) -> Result<(), String> {
        if !self.is_initiated() {
            return Err("Ingen vedlikehold er initiert".to_string());
        }

        let mut inner = self.inner.lock().await;

        // Avbryt bakgrunnsoppgaven
        if let Some(handle) = inner.abort_handle.take() {
            handle.abort();
        }

        // Slett varselet fra PG
        if let Some(nid) = inner.announcement_node_id.take() {
            if let Err(e) = sqlx::query("DELETE FROM nodes WHERE id = $1")
                .bind(nid)
                .execute(db)
                .await
            {
                tracing::warn!("Kunne ikke slette varsel fra PG: {e}");
            }
        }

        inner.scheduled_at = None;
        inner.initiated_by = None;

        self.initiated.store(false, Ordering::Relaxed);
        self.active.store(false, Ordering::Relaxed);

        tracing::info!("Vedlikehold avbrutt");
        Ok(())
    }
}

/// Hent kjørende jobber fra job_queue.
async fn fetch_running_jobs(db: &PgPool) -> Result<Vec<RunningJob>, sqlx::Error> {
    let rows = sqlx::query_as::<_, (Uuid, String, Option<chrono::DateTime<Utc>>, Option<Uuid>)>(
        "SELECT id, job_type, started_at, collection_node_id FROM job_queue WHERE status = 'running'"
    )
    .fetch_all(db)
    .await?;

    Ok(rows.into_iter().map(|(id, job_type, started_at, collection_node_id)| {
        RunningJob {
            id,
            job_type,
            started_at: started_at.map(|dt| dt.to_rfc3339()),
            collection_node_id,
        }
    }).collect())
}

/// Bakgrunnsoppgave som koordinerer nedstengningen.
async fn shutdown_coordinator(
    state: MaintenanceState,
    db: PgPool,
    scheduled_at: DateTime<Utc>,
    announcement_id: Uuid,
) {
    // Vent til vedlikeholdstidspunkt
    let wait_duration = (scheduled_at - Utc::now()).to_std().unwrap_or_default();
    if !wait_duration.is_zero() {
        tracing::info!(
            seconds = wait_duration.as_secs(),
            "Venter til vedlikeholdstidspunkt"
        );
        tokio::time::sleep(wait_duration).await;
    }

    // Aktiver vedlikeholdsmodus
    state.active.store(true, Ordering::Relaxed);
    tracing::warn!("Vedlikeholdsmodus AKTIV — nye jobber og LiveKit-rom blokkert");

    // Oppdater varselet til å reflektere at vedlikehold er i gang
    let active_meta = serde_json::json!({
        "announcement_type": "critical",
        "scheduled_at": scheduled_at.to_rfc3339(),
        "blocks_new_sessions": true,
        "maintenance_shutdown": true,
        "maintenance_active": true,
    });
    let _ = sqlx::query(
        "UPDATE nodes SET title = $2, content = $3, metadata = $4 WHERE id = $1",
    )
    .bind(announcement_id)
    .bind("Vedlikehold pågår")
    .bind("Systemet stenger ned. Vent til vedlikeholdet er ferdig.")
    .bind(&active_meta)
    .execute(&db)
    .await;

    // Vent på at kjørende jobber fullføres (maks 5 minutter)
    let timeout = std::time::Duration::from_secs(300);
    let start = std::time::Instant::now();

    loop {
        match fetch_running_jobs(&db).await {
            Ok(jobs) if jobs.is_empty() => {
                tracing::info!("Ingen kjørende jobber — klar for restart");
                break;
            }
            Ok(jobs) => {
                tracing::info!(
                    count = jobs.len(),
                    "Venter på {} kjørende jobber",
                    jobs.len()
                );
            }
            Err(e) => {
                tracing::error!("Feil ved sjekk av kjørende jobber: {e}");
            }
        }

        if start.elapsed() > timeout {
            tracing::warn!("Timeout (5 min) — tvinger nedstengning med kjørende jobber");
            break;
        }

        tokio::time::sleep(std::time::Duration::from_secs(5)).await;
    }

    // Slett varselet
    if let Err(e) = sqlx::query("DELETE FROM nodes WHERE id = $1")
        .bind(announcement_id)
        .execute(&db)
        .await
    {
        tracing::warn!("Kunne ikke slette varsel fra PG: {e}");
    }

    tracing::warn!("Avslutter prosessen for vedlikehold — systemd vil restarte");

    // Gi litt tid til at siste logglinjer skrives
    tokio::time::sleep(std::time::Duration::from_millis(500)).await;

    // Avslutt prosessen — systemd vil restarte maskinrommet
    std::process::exit(0);
}