launcher/src/entity_cache.rs

//! Process-wide liveness cache for HA / RouterOS entity states.
//!
//! The dashboard's fleet probe runs every 30 s. Without a cache, every tick
//! issues one HA `GET /api/states/...` per node/extra and — worse — a full
//! RouterOS API login per switch. A dead switch must not be hammered, and a
//! live one must not be asked to authenticate every 30 s. The cache keeps a
//! state for a TTL (HA: 120 s, RouterOS: 300 s); a power action invalidates it
//! so the next probe sees the real result immediately.

use std::collections::HashMap;
use std::sync::{LazyLock, Mutex};
use std::time::{Duration, Instant};

#[derive(Clone, Copy)]
struct Entry {
    value: Option<bool>,
    checked: Instant,
    ttl: Duration,
}

static CACHE: LazyLock<Mutex<HashMap<String, Entry>>> =
    LazyLock::new(|| Mutex::new(HashMap::new()));

fn lock_cache() -> std::sync::MutexGuard<'static, HashMap<String, Entry>> {
    CACHE
        .lock()
        .unwrap_or_else(std::sync::PoisonError::into_inner)
}

/// Fresh cached state (`true`/`false`) or `None` when absent/expired.
pub fn fresh(key: &str) -> Option<Option<bool>> {
    let mut cache = lock_cache();
    match cache.get(key) {
        Some(e) if e.checked.elapsed() < e.ttl => Some(e.value),
        _ => {
            cache.remove(key);
            None
        }
    }
}

/// Store a state for `ttl`.
pub fn store(key: &str, value: Option<bool>, ttl: Duration) {
    let mut cache = lock_cache();
    cache.insert(
        key.to_string(),
        Entry {
            value,
            checked: Instant::now(),
            ttl,
        },
    );
    // Keep the map bounded by dropping expired entries on write.
    cache.retain(|_, e| e.checked.elapsed() < e.ttl);
}

/// Drop one cached entity (e.g. after a HA/ RouterOS toggle).
pub fn invalidate(key: &str) {
    lock_cache().remove(key);
}

/// Drop every entity (after a power flow; the dashboard's next tick re-probes).
pub fn invalidate_all() {
    lock_cache().clear();
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn fresh_store_and_expire() {
        store("ha:switch.x", Some(true), Duration::from_secs(2));
        assert_eq!(fresh("ha:switch.x"), Some(Some(true)));
        // A zero TTL means "do not trust this entry".
        store("ha:switch.y", Some(false), Duration::ZERO);
        assert_eq!(fresh("ha:switch.y"), None);
    }

    #[test]
    fn invalidation_forces_reprobe() {
        store(
            "mikrotik:sw:10.0.0.1",
            Some(false),
            Duration::from_secs(300),
        );
        assert_eq!(fresh("mikrotik:sw:10.0.0.1"), Some(Some(false)));
        invalidate("mikrotik:sw:10.0.0.1");
        assert_eq!(fresh("mikrotik:sw:10.0.0.1"), None);
        store("ha:switch.z", Some(true), Duration::from_secs(120));
        invalidate_all();
        assert_eq!(fresh("ha:switch.z"), None);
    }
}