osdns 0.1.1

use std::collections::HashMap;
use std::fmt;
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::mpsc;
use std::sync::{Arc, Mutex};
use std::thread;
use std::time::{Duration, Instant};

use crate::capability::BackendKind;
use crate::error::{Error, Result};
use crate::ownership::ResourceId;

/// An observed change to a DNS resource.
///
/// Variants are `#[non_exhaustive]` so new event kinds can be added without a
/// breaking change. Match with a wildcard arm.
///
/// Watch callbacks must only enqueue or coalesce events; expensive or
/// mutating logic must never run inside a callback. Events caused by our own
/// mutations are suppressed from the user callback path (under
/// [`ConflictPolicy::Enforce`](crate::ConflictPolicy) the reconciler still
/// observes them via read-back before suppression applies).
#[derive(Debug, Clone, PartialEq, Eq)]
#[non_exhaustive]
pub enum DnsEvent {
    /// The resource's DNS configuration changed.
    ResourceChanged {
        /// The resource that changed.
        resource: ResourceId,
    },
    /// The resource disappeared (e.g. an interface was removed).
    ResourceRemoved {
        /// The resource that disappeared.
        resource: ResourceId,
    },
}

impl DnsEvent {
    pub(crate) fn resource(&self) -> &ResourceId {
        match self {
            DnsEvent::ResourceChanged { resource } => resource,
            DnsEvent::ResourceRemoved { resource } => resource,
        }
    }

    fn into_owned(self) -> (ResourceId, bool) {
        match self {
            DnsEvent::ResourceChanged { resource } => (resource, false),
            DnsEvent::ResourceRemoved { resource } => (resource, true),
        }
    }
}

/// Callback invoked by a backend's event thread.
///
/// Must be `Send + Sync` because it runs on a native watcher thread. Keep it
/// short: enqueue the event and return. Never call [`DnsManager::apply`](crate::DnsManager::apply),
/// [`Lease::restore`](crate::Lease::restore), or other mutating APIs from
/// inside the callback; doing so risks deadlock with the coalescer and
/// reconciler threads.
pub type WatchCallback = Arc<dyn Fn(&DnsEvent) + Send + Sync>;

/// A live watch. Cancels the underlying native notification when dropped or
/// explicitly stopped.
///
/// `stop` consumes the handle; dropping has the same effect. Stopping is
/// idempotent and never fails. Stopping a public watch never disables
/// [`ConflictPolicy::Enforce`](crate::ConflictPolicy): Enforce keeps its own
/// internal observation alive while leases are active.
///
/// ```no_run
/// # use osdns::DnsManager;
/// # use std::sync::Arc;
/// # fn main() -> osdns::Result<()> {
/// # let manager = DnsManager::builder().owner("io.example.agent").build()?;
/// let watch = manager.watch(Arc::new(|event| println!("{event:?}")))?;
/// // ... observe events ...
/// watch.stop();
/// # Ok(())
/// # }
/// ```
pub struct WatchHandle {
    flag: Arc<AtomicBool>,
    cancel: Mutex<Option<Box<dyn FnOnce() + Send>>>,
}

impl WatchHandle {
    #[allow(dead_code)]
    pub(crate) fn new(flag: Arc<AtomicBool>, cancel: impl FnOnce() + Send + 'static) -> Self {
        Self {
            flag,
            cancel: Mutex::new(Some(Box::new(cancel))),
        }
    }

    pub(crate) fn is_active(&self) -> bool {
        !self.flag.load(Ordering::Acquire)
    }

    /// Stops the watch and releases the underlying native notification.
    ///
    /// Consumes the handle; equivalent to dropping it. After `stop` no
    /// further callbacks will be delivered for this watch.
    pub fn stop(mut self) {
        self.deactivate();
    }

    fn deactivate(&mut self) {
        self.flag.store(true, Ordering::Release);
        if let Some(cancel) = self
            .cancel
            .lock()
            .unwrap_or_else(|poisoned| poisoned.into_inner())
            .take()
        {
            cancel();
        }
    }
}

impl Drop for WatchHandle {
    fn drop(&mut self) {
        self.deactivate();
    }
}

impl fmt::Debug for WatchHandle {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_struct("WatchHandle")
            .field("active", &self.is_active())
            .finish()
    }
}

/// Tracks the resources the transaction engine has recently mutated, so
/// watcher events generated by our own mutations are suppressed instead of
/// being reported as external changes.
#[derive(Debug, Default)]
pub(crate) struct SuppressionRegistry {
    entries: Mutex<HashMap<ResourceId, Instant>>,
}

impl SuppressionRegistry {
    const WINDOW: Duration = Duration::from_millis(500);

    pub(crate) fn new() -> Self {
        Self::default()
    }

    pub(crate) fn suppress(&self, resource: &ResourceId) {
        let mut entries = self
            .entries
            .lock()
            .unwrap_or_else(|poisoned| poisoned.into_inner());
        Self::prune(&mut entries);
        entries.insert(resource.clone(), Instant::now());
    }

    pub(crate) fn is_suppressed(&self, resource: &ResourceId) -> bool {
        let mut entries = self
            .entries
            .lock()
            .unwrap_or_else(|poisoned| poisoned.into_inner());
        Self::prune(&mut entries);
        entries.contains_key(resource)
    }

    fn prune(entries: &mut HashMap<ResourceId, Instant>) {
        let now = Instant::now();
        entries.retain(|_, suppressed_at| now.duration_since(*suppressed_at) < Self::WINDOW);
    }
}

/// Wraps a callback so bursts of events for the same resource coalesce into
/// a single delivery.
///
/// The returned callback never blocks: it enqueues into a channel drained by
/// a dedicated thread that blocks on `recv` when idle (zero polling) and
/// flushes once the stream stays quiet for `window`.
pub(crate) fn spawn_coalescer(
    kind: BackendKind,
    callback: WatchCallback,
    window: Duration,
) -> Result<WatchCallback> {
    let (tx, rx) = mpsc::channel::<DnsEvent>();
    thread::Builder::new()
        .name("osdns-watch-coalescer".to_string())
        .spawn(move || {
            let mut pending: HashMap<ResourceId, DnsEvent> = HashMap::new();
            while let Ok(first) = rx.recv() {
                let resource = first.resource().clone();
                pending.insert(resource, first);
                let deadline = Instant::now() + window;
                loop {
                    match rx.recv_timeout(deadline.saturating_duration_since(Instant::now())) {
                        Ok(event) => {
                            let (resource, removed) = event.clone().into_owned();
                            if removed {
                                pending.insert(resource, event);
                            } else {
                                pending.entry(resource).or_insert(event);
                            }
                        }
                        Err(mpsc::RecvTimeoutError::Timeout) => break,
                        Err(mpsc::RecvTimeoutError::Disconnected) => {
                            flush(&callback, &mut pending);
                            return;
                        }
                    }
                }
                flush(&callback, &mut pending);
            }
        })
        .map_err(|e| Error::Platform {
            backend: kind,
            message: format!("cannot spawn coalescer thread: {e}"),
        })?;
    Ok(Arc::new(move |event| {
        let _ = tx.send(event.clone());
    }))
}

fn flush(callback: &WatchCallback, pending: &mut HashMap<ResourceId, DnsEvent>) {
    for event in pending.values() {
        callback(event);
    }
    pending.clear();
}