vcon_server/

serve.rs

//! `vcon-server serve`: the server, run from its configuration.

use std::fmt;
use std::future::Future;
use std::io;
use std::sync::Arc;
use std::time::Duration;

use tokio::net::TcpListener;

use crate::api::{self, App, Limits, TlsError};
use crate::clock::{Clock, SystemClock};
use crate::config::Config;
use crate::datadir::{self, LockError, Work};
use crate::envelope::{Key, KeyError};
use crate::fs::OsFs;
use crate::retention::Retention;
use crate::tenants::{KeyOrder, RotationBacklog, TenantError, Tenants};
use time::OffsetDateTime;

/// Why the server did not start, or stopped.
#[derive(Debug)]
pub enum ServeError {
    /// The data directory could not be taken.
    DataDir(LockError),
    /// The master key file could not be read.
    MasterKey(KeyError),
    /// server.sqlite was made under another master key.
    WrongMasterKey,
    /// server.sqlite could not be opened.
    Registry(TenantError),
    /// Some tenant's data is not under its current key.
    KeysOutOfOrder(KeyOrder),
    /// The certificate or key could not be used.
    Tls(TlsError),
    /// The listening address could not be bound.
    Listen(io::Error),
    /// Serving failed.
    Serve(io::Error),
}

impl fmt::Display for ServeError {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            ServeError::DataDir(e) => write!(f, "{e}"),
            ServeError::MasterKey(e) => write!(f, "the master key cannot be read: {e:?}"),
            ServeError::WrongMasterKey => {
                f.write_str("the master key is not the one server.sqlite was made under")
            }
            ServeError::Registry(e) => write!(f, "server.sqlite cannot be opened: {e:?}"),
            ServeError::KeysOutOfOrder(order) => {
                let mut parts = Vec::new();
                if order.wrapped_elsewhere > 0 {
                    parts.push(format!(
                        "{} tenant key(s) do not unwrap under the master key",
                        order.wrapped_elsewhere
                    ));
                }
                if order.rotations_stopped > 0 {
                    parts.push(format!(
                        "{} key rotation(s) stopped part way; run rekey tenant again",
                        order.rotations_stopped
                    ));
                }
                if order.data_under_other_keys > 0 {
                    parts.push(format!(
                        "{} tenant(s) with data under another key",
                        order.data_under_other_keys
                    ));
                }
                write!(
                    f,
                    "{}; vcon-server check names the tenants",
                    parts.join("; ")
                )
            }
            ServeError::Tls(e) => write!(f, "the TLS certificate or key cannot be used: {e:?}"),
            ServeError::Listen(e) => write!(f, "the listening address cannot be bound: {e}"),
            ServeError::Serve(e) => write!(f, "serving failed: {e}"),
        }
    }
}

/// Runs the server until `shutdown` completes: locks the data directory,
/// opens the registry under the master key, and serves the API over TLS.
pub async fn run(config: &Config, shutdown: impl Future<Output = ()>) -> Result<(), ServeError> {
    let _lock =
        datadir::lock(&config.data_dir, Work::Serve.sharing()).map_err(ServeError::DataDir)?;
    let master = Key::load_master(&config.master_key).map_err(ServeError::MasterKey)?;
    let tenants = Tenants::open(
        &config.data_dir,
        master,
        Arc::new(OsFs),
        Arc::new(SystemClock),
    )
    .map_err(|e| match e {
        TenantError::WrongKey => ServeError::WrongMasterKey,
        other => ServeError::Registry(other),
    })?;
    let order = tenants.key_order().map_err(ServeError::Registry)?;
    if !order.is_sound() {
        return Err(ServeError::KeysOutOfOrder(order));
    }
    let tls = api::tls_config(&config.tls_cert, &config.tls_key).map_err(ServeError::Tls)?;
    let listener = TcpListener::bind(config.listen)
        .await
        .map_err(ServeError::Listen)?;
    let tenants = Arc::new(tenants);
    // The purge sweep: at start, then purge_interval after each sweep ends,
    // until the server stops.
    let (stop, stopped) = tokio::sync::oneshot::channel::<()>();
    let (swept, default) = (tenants.clone(), config.retention_default);
    let sweeper = tokio::spawn(every(
        config.purge_interval,
        async {
            let _ = stopped.await;
        },
        move || {
            let tenants = swept.clone();
            async move {
                let _ = tokio::task::spawn_blocking(move || sweep(&tenants, default)).await;
            }
        },
    ));
    // The rotation warning: at start, then each day.
    let (stop_warning, warning_stopped) = tokio::sync::oneshot::channel::<()>();
    let (watched, rotation_warn) = (tenants.clone(), config.rotation_warn);
    let warner = tokio::spawn(every(
        ROTATION_CHECK,
        async {
            let _ = warning_stopped.await;
        },
        move || {
            let tenants = watched.clone();
            async move {
                let _ = tokio::task::spawn_blocking(move || {
                    warn_if_rotation_overdue(&tenants, rotation_warn);
                })
                .await;
            }
        },
    ));
    let app = app_from(config, tenants);
    tracing::info!(listen = %config.listen, "serving");
    let served = api::serve(listener, tls, api::router(Arc::new(app)), shutdown).await;
    let _ = stop.send(());
    let _ = stop_warning.send(());
    let _ = sweeper.await;
    let _ = warner.await;
    served.map_err(ServeError::Serve)?;
    tracing::info!("stopped");
    Ok(())
}

/// What one purge sweep did.
#[derive(Debug, Default, Clone, Copy, PartialEq, Eq)]
pub struct SweepReport {
    /// Versions purged, over every tenant.
    pub purged: u64,
    /// Tenants that could not be swept; they wait for the next sweep.
    pub failed: u64,
}

/// One purge sweep: each tenant purges what is due under its own retention
/// period, else `default`. A tenant that cannot be swept, such as one whose
/// period does not read back or that is being removed, is counted and left
/// for the next sweep. The sweep is logged with counts only, never a tenant.
pub fn sweep(tenants: &Tenants, default: Retention) -> SweepReport {
    let mut report = SweepReport::default();
    match tenants.ids() {
        Ok(ids) => {
            for id in ids {
                let purged = tenants.retention(id).and_then(|own| {
                    let store = tenants.store(id)?;
                    Ok(store.purge_due(own.unwrap_or(default)))
                });
                match purged {
                    Ok(Ok(done)) => report.purged += done.purged,
                    _ => report.failed += 1,
                }
            }
        }
        Err(_) => report.failed += 1,
    }
    tracing::info!(
        purged = report.purged,
        failed = report.failed,
        "purge sweep"
    );
    report
}

/// Runs `job` at once, then again `interval` after each run ends, until
/// `shutdown` completes; a run still going when it does is not awaited.
pub async fn every<F, Fut>(interval: Duration, shutdown: impl Future<Output = ()>, mut job: F)
where
    F: FnMut() -> Fut,
    Fut: Future<Output = ()>,
{
    tokio::pin!(shutdown);
    loop {
        tokio::select! {
            biased;
            () = &mut shutdown => return,
            () = job() => {}
        }
        tokio::select! {
            biased;
            () = &mut shutdown => return,
            () = tokio::time::sleep(interval) => {}
        }
    }
}

/// What the API serves from, as the configuration sets it.
fn app_from(config: &Config, tenants: Arc<Tenants>) -> App {
    App {
        tenants,
        limits: Limits {
            body: config.body_limit,
            inflated: config.inflated_limit,
            body_timeout: config.read_timeout,
        },
        supported: config.supported_critical.clone(),
        retention_default: config.retention_default,
    }
}

/// Whether key rotation is overdue: the oldest purge or removal waiting
/// for it is older than `rotation_warn`.
pub fn rotation_overdue(
    backlog: &RotationBacklog,
    rotation_warn: Duration,
    now: OffsetDateTime,
) -> bool {
    [backlog.oldest_purge, backlog.oldest_removal]
        .into_iter()
        .flatten()
        .any(|at| now - at > rotation_warn)
}

/// How often `serve` looks at the rotation backlog.
const ROTATION_CHECK: Duration = Duration::from_secs(86_400);

/// Logs a warning, with counts only, when key rotation is overdue.
fn warn_if_rotation_overdue(tenants: &Tenants, rotation_warn: Duration) {
    match tenants.rotation_backlog() {
        Ok(backlog) if rotation_overdue(&backlog, rotation_warn, SystemClock.now()) => {
            tracing::warn!(
                tenants_with_purges = backlog.tenants_with_purges,
                purges = backlog.purges,
                removals = backlog.removals,
                "key rotation overdue: purged vCons and removed tenants stay readable on old copies until it"
            );
        }
        Ok(_) => {}
        Err(_) => tracing::warn!("the key rotation backlog could not be read"),
    }
}

#[cfg(test)]
mod tests {
    use std::time::Duration;

    use super::*;

    mod startup {
        use time::macros::datetime;

        use super::*;

        const NOW: OffsetDateTime = datetime!(2026-10-25 12:00 UTC);
        const DAY: Duration = Duration::from_secs(86_400);

        #[test]
        fn rotation_is_overdue_once_the_oldest_waiting_work_passes_rotation_warn() {
            let backlog =
                |purge: Option<OffsetDateTime>, removal: Option<OffsetDateTime>| RotationBacklog {
                    tenants_with_purges: u64::from(purge.is_some()),
                    purges: u64::from(purge.is_some()),
                    oldest_purge: purge,
                    removals: u64::from(removal.is_some()),
                    oldest_removal: removal,
                };
            let day_ago = Some(datetime!(2026-10-24 12:00 UTC));
            let older = Some(datetime!(2026-10-24 11:59:59 UTC));
            for (b, overdue) in [
                (backlog(None, None), false),
                (backlog(day_ago, None), false),
                (backlog(older, None), true),
                (backlog(None, older), true),
                (backlog(day_ago, older), true),
            ] {
                assert_eq!(rotation_overdue(&b, DAY, NOW), overdue, "{b:?}");
            }
        }

        #[test]
        fn a_refusal_for_keys_out_of_order_counts_and_points_at_check() {
            let order = KeyOrder {
                wrapped_elsewhere: 1,
                rotations_stopped: 2,
                data_under_other_keys: 3,
            };
            let text = ServeError::KeysOutOfOrder(order).to_string();
            for words in [
                "1 tenant key",
                "2 key rotation",
                "3 tenant",
                "vcon-server check",
            ] {
                assert!(text.contains(words), "{words}: {text}");
            }
        }
    }
    use crate::retention::Retention;

    #[test]
    fn the_api_takes_its_limits_and_extensions_from_the_configuration() {
        let dir = tempfile::tempdir().expect("temp dir");
        let tenants = Arc::new(
            Tenants::open(
                dir.path(),
                Key::generate().expect("key"),
                Arc::new(OsFs),
                Arc::new(SystemClock),
            )
            .expect("opens"),
        );
        let config = Config {
            listen: "127.0.0.1:8443".parse().expect("address"),
            tls_cert: "cert.pem".into(),
            tls_key: "key.pem".into(),
            data_dir: dir.path().to_owned(),
            master_key: "master.key".into(),
            retention_default: Retention::Months(18),
            purge_interval: Duration::from_secs(3600),
            rotation_warn: Duration::from_secs(86400),
            body_limit: 12_345,
            inflated_limit: 67_890,
            read_timeout: Duration::from_secs(17),
            supported_critical: ["known".to_owned()].into(),
        };
        let app = app_from(&config, tenants.clone());
        assert!(Arc::ptr_eq(&app.tenants, &tenants));
        assert_eq!(app.limits.body, 12_345);
        assert_eq!(app.limits.inflated, 67_890);
        assert_eq!(app.limits.body_timeout, Duration::from_secs(17));
        assert_eq!(app.supported, config.supported_critical);
        assert_eq!(app.retention_default, config.retention_default);
    }

    mod sweeps {
        use std::sync::atomic::{AtomicUsize, Ordering};
        use std::time::Duration;

        use time::macros::datetime;
        use uuid::Uuid;

        use super::super::*;
        use crate::audit::Actor;
        use crate::ids::{KeyId, TenantId};
        use crate::retention::Retention;
        use crate::store::Caller;
        use crate::store::testing::vcon;
        use crate::tenants::fixture;

        const CALLER: Caller = Caller { key: KeyId([7; 8]) };

        /// A vCon stored and deleted in `id` while seven years apply, so
        /// it waits hidden for a sweep.
        fn deleted(tenants: &Tenants, id: TenantId, uuid: &str) -> Uuid {
            let store = tenants.store(id).expect("its store");
            let (found, body) = vcon(uuid, "kept");
            store.create(&CALLER, &found, &body).expect("stored");
            store
                .delete(&CALLER, found.uuid, Retention::Years(7))
                .expect("deleted");
            found.uuid
        }

        /// How many versions `id` still keeps.
        fn kept(tenants: &Tenants, id: TenantId) -> u64 {
            tenants
                .store(id)
                .expect("its store")
                .counts()
                .expect("counts")
                .versions
        }

        #[test]
        fn a_sweep_purges_what_is_due_in_each_tenant_under_its_own_period() {
            let f = fixture();
            let own = f
                .tenants
                .create(Actor::Cli, "own", Some(Retention::Days(1)))
                .expect("made");
            let default = f.tenants.create(Actor::Cli, "default", None).expect("made");
            let long = f
                .tenants
                .create(Actor::Cli, "long", Some(Retention::Years(7)))
                .expect("made");
            deleted(&f.tenants, own, "0199e5c6-0d7a-7c11-8000-000000000001");
            deleted(&f.tenants, default, "0199e5c6-0d7a-7c11-8000-000000000002");
            deleted(&f.tenants, long, "0199e5c6-0d7a-7c11-8000-000000000003");
            f.clock.set(datetime!(2026-10-25 13:00 UTC));

            let report = sweep(&f.tenants, Retention::Days(1));

            assert_eq!(
                report,
                SweepReport {
                    purged: 2,
                    failed: 0
                }
            );
            assert_eq!(kept(&f.tenants, own), 0);
            assert_eq!(kept(&f.tenants, default), 0);
            assert_eq!(kept(&f.tenants, long), 1, "seven years have not passed");
        }

        #[test]
        fn a_sweep_goes_on_past_a_tenant_it_cannot_sweep() {
            let f = fixture();
            let damaged = f.tenants.create(Actor::Cli, "damaged", None).expect("made");
            let fine = f.tenants.create(Actor::Cli, "fine", None).expect("made");
            deleted(&f.tenants, damaged, "0199e5c6-0d7a-7c11-8000-000000000001");
            deleted(&f.tenants, fine, "0199e5c6-0d7a-7c11-8000-000000000002");
            f.tenants
                .conn_for_test()
                .execute(
                    "UPDATE tenant SET retention = '7w' WHERE id = ?1",
                    [damaged.0],
                )
                .expect("damaged");
            f.clock.set(datetime!(2026-10-25 13:00 UTC));

            let report = sweep(&f.tenants, Retention::Days(1));

            assert_eq!(
                report,
                SweepReport {
                    purged: 1,
                    failed: 1
                }
            );
            assert_eq!(kept(&f.tenants, fine), 0);
            assert_eq!(kept(&f.tenants, damaged), 1, "left for a later sweep");
        }

        #[tokio::test(start_paused = true)]
        async fn every_runs_at_start_then_after_each_interval_until_shutdown() {
            let runs = Arc::new(AtomicUsize::new(0));
            let (stop, stopped) = tokio::sync::oneshot::channel::<()>();
            let counted = runs.clone();
            let task = tokio::spawn(every(
                Duration::from_secs(60),
                async move {
                    let _ = stopped.await;
                },
                move || {
                    counted.fetch_add(1, Ordering::SeqCst);
                    async {}
                },
            ));
            let settle = || async {
                for _ in 0..10 {
                    tokio::task::yield_now().await;
                }
            };

            settle().await;
            assert_eq!(runs.load(Ordering::SeqCst), 1, "at start");
            tokio::time::advance(Duration::from_secs(59)).await;
            settle().await;
            assert_eq!(runs.load(Ordering::SeqCst), 1, "not before the interval");
            tokio::time::advance(Duration::from_secs(1)).await;
            settle().await;
            assert_eq!(runs.load(Ordering::SeqCst), 2, "after the interval");

            stop.send(()).expect("still running");
            task.await.expect("ends on shutdown");
            tokio::time::advance(Duration::from_secs(600)).await;
            assert_eq!(runs.load(Ordering::SeqCst), 2, "none after shutdown");
        }
    }
}