imks/etcd.rs

use etcd_client::{Client, PutOptions, GetOptions, WatchOptions};
use std::sync::Arc;
use tokio::sync::Mutex;
use tokio_stream::StreamExt;

/// etcd-backed config reader. Priority: etcd > env var > default.
pub struct EtcdConfig {
    client: Arc<Mutex<Client>>,
    prefix: String,
}

impl EtcdConfig {
    pub async fn connect(endpoints: Vec<String>, prefix: &str) -> Result<Self, String> {
        let client = Client::connect(endpoints, None)
            .await
            .map_err(|e| format!("etcd connect: {e}"))?;
        Ok(Self { client: Arc::new(Mutex::new(client)), prefix: prefix.to_string() })
    }

    /// Get config value: etcd first, then env var, then default.
    pub async fn get(&self, key: &str, default: &str) -> String {
        let etcd_key = format!("{}config/{}", self.prefix, key);
        if let Ok(mut c) = self.client.try_lock() {
            if let Ok(resp) = c.get(etcd_key.as_str(), None).await {
                if let Some(kv) = resp.kvs().first() {
                    if let Ok(v) = kv.value_str() {
                        if !v.is_empty() {
                            tracing::info!(key, value = v, "config from etcd");
                            return v.to_string();
                        }
                    }
                }
            }
        }
        if let Ok(v) = std::env::var(key) {
            if !v.is_empty() {
                return v;
            }
        }
        default.to_string()
    }

    pub async fn get_parsed<T: std::str::FromStr>(&self, key: &str, default: T) -> T
    where T::Err: std::fmt::Display, T: std::fmt::Display
    {
        let s = self.get(key, &default.to_string()).await;
        s.parse().unwrap_or(default)
    }

    /// Get the etcd client for use by ServiceRegistry.
    pub fn client(&self) -> Arc<Mutex<Client>> {
        self.client.clone()
    }

    /// Discover service instances registered under /{prefix}/services/{service_name}/.
    pub async fn discover_service(&self, service_name: &str) -> Result<Vec<String>, String> {
        let prefix = format!("{}services/{}/", self.prefix, service_name);
        let mut client = self.client.lock().await;
        let resp = client
            .get(prefix.as_str(), Some(GetOptions::new().with_prefix()))
            .await
            .map_err(|e| format!("etcd get {prefix}: {e}"))?;

        let mut addrs = Vec::new();
        for kv in resp.kvs() {
            if let Ok(value) = kv.value_str() {
                if let Ok(instance) = serde_json::from_str::<serde_json::Value>(value) {
                    if let Some(addr) = instance.get("addr").and_then(|v| v.as_str()) {
                        addrs.push(addr.to_string());
                    }
                }
            }
        }
        tracing::info!(service = service_name, count = addrs.len(), "discovered instances");
        Ok(addrs)
    }

    /// Watch a service for live updates.
    pub fn start_service_watcher(&self, service_name: &str) {
        let client = self.client.clone();
        let prefix = self.prefix.clone();
        let svc = service_name.to_string();
        let watch_prefix = format!("{}services/{}/", prefix, svc);

        tokio::spawn(async move {
            loop {
                let mut stream = {
                    let mut c = client.lock().await;
                    match c.watch(watch_prefix.as_str(), Some(WatchOptions::new().with_prefix())).await {
                        Ok(s) => s,
                        Err(e) => {
                            tracing::warn!(service = %svc, error = %e, "watch failed, retry in 3s");
                            tokio::time::sleep(std::time::Duration::from_secs(3)).await;
                            continue;
                        }
                    }
                };
                while let Some(resp) = stream.next().await {
                    if let Ok(resp) = resp {
                        for event in resp.events() {
                            if let Some(kv) = event.kv() {
                                let addr = kv.value_str().unwrap_or_default();
                                let key = kv.key_str().unwrap_or_default();
                                match event.event_type() {
                                    etcd_client::EventType::Put => {
                                        tracing::info!(service = %svc, key, addr, "service up");
                                    }
                                    etcd_client::EventType::Delete => {
                                        tracing::info!(service = %svc, key, "service down");
                                    }
                                }
                            }
                        }
                    }
                }
            }
        });
    }
}

/// Register this service instance in etcd.
pub struct ServiceRegistry {
    client: Arc<Mutex<Client>>,
    prefix: String,
}

impl ServiceRegistry {
    pub fn new(client: Arc<Mutex<Client>>, prefix: &str) -> Self {
        Self { client, prefix: prefix.to_string() }
    }

    pub async fn register(&self, service_name: &str, addr: &str) -> Result<(), String> {
        let instance_id = uuid::Uuid::now_v7().to_string();
        let addr = addr.to_string();
        let key = format!("{}services/{}/{}", self.prefix, service_name, instance_id);

        let instance = serde_json::json!({
            "addr": &addr,
            "port": 0,
            "version": env!("CARGO_PKG_VERSION"),
        });
        let value = serde_json::to_string(&instance).map_err(|e| format!("json: {e}"))?;

        let lease = {
            let mut client = self.client.lock().await;
            client.lease_grant(15, None).await.map_err(|e| format!("lease: {e}"))?
        };

        {
            let mut client = self.client.lock().await;
            let opts = PutOptions::new().with_lease(lease.id());
            client.put(key.clone(), value, Some(opts)).await.map_err(|e| format!("put: {e}"))?;
        }

        tracing::info!(service = service_name, instance = %instance_id, addr = %addr, "registered in etcd");

        let c = self.client.clone();
        tokio::spawn(async move {
            loop {
                let result = {
                    let mut client = c.lock().await;
                    client.lease_keep_alive(lease.id()).await
                };
                match result {
                    Ok((_keeper, mut stream)) => {
                        while stream.next().await.is_some() {}
                    }
                    Err(e) => tracing::warn!(lease_id = lease.id(), error = %e, "keepalive failed"),
                }
                tokio::time::sleep(std::time::Duration::from_secs(5)).await;
                let new_lease = {
                    let mut client = c.lock().await;
                    client.lease_grant(15, None).await
                };
                if let Ok(lr) = new_lease {
                    let instance = serde_json::json!({"addr": &addr, "port": 0, "version": env!("CARGO_PKG_VERSION")});
                    if let Ok(v) = serde_json::to_string(&instance) {
                        let mut client = c.lock().await;
                        let opts = PutOptions::new().with_lease(lr.id());
                        let _ = client.put(key.clone(), v, Some(opts)).await;
                    }
                }
            }
        });

        Ok(())
    }
}