node: add server election

peer_mgr

server: add peer_mgr

peer_mgr
This commit is contained in:
Luna Yao
2026-02-23 00:06:46 +01:00
parent 238858ebab
commit 57a2428f21
4 changed files with 73 additions and 30 deletions
+55 -15
View File
@@ -1,28 +1,30 @@
use crate::common::global_ctx::GlobalCtxEvent;
use crate::common::PeerId;
use crate::dns::config::DNS_SERVER_RPC_ADDR;
use crate::dns::config::{DNS_SERVER_ELECTION_INTERVAL, DNS_SERVER_RPC_ADDR};
use crate::dns::peer_mgr::DnsPeerMgr;
use crate::dns::server::DnsServer;
use crate::peers::peer_manager::PeerManager;
use crate::peers::NicPacketFilter;
use crate::proto::dns::{DnsNodeMgrRpcClientFactory, DnsPeerMgrRpcServer, HeartbeatRequest};
use crate::proto::rpc_impl::standalone::StandAloneClient;
use crate::proto::rpc_impl::standalone::{StandAloneClient, StandAloneServer};
use crate::proto::rpc_types::controller::BaseController;
use crate::tunnel::tcp::TcpTunnelConnector;
use crate::tunnel::tcp::{TcpTunnelConnector, TcpTunnelListener};
use std::sync::Arc;
use std::time::Duration;
use tokio::sync::{broadcast, Notify};
use tokio::task::JoinSet;
use tokio::time::{sleep_until, Instant};
use tokio::time::{sleep, sleep_until, Instant};
use uuid::Uuid;
#[derive(Debug)]
pub struct DnsNode {
mgr: Arc<DnsPeerMgr>,
election: Arc<Notify>,
peer_mgr: Arc<PeerManager>,
}
impl DnsNode {
pub fn new(peer_mgr: Arc<PeerManager>, election: Arc<Notify>) -> Self {
pub fn new(peer_mgr: Arc<PeerManager>) -> Self {
let mgr = Arc::new(DnsPeerMgr::new(peer_mgr.clone()));
peer_mgr
.get_peer_rpc_mgr()
@@ -33,17 +35,55 @@ impl DnsNode {
&peer_mgr.get_global_ctx_ref().get_network_name(),
);
Self {
mgr,
election,
}
Self { mgr, peer_mgr }
}
pub fn id(&self) -> Uuid {
self.mgr.get_global_ctx_ref().get_id()
self.peer_mgr.get_global_ctx_ref().get_id()
}
pub async fn run(&self) {
let election = Notify::new();
tokio::join!(self.run_election(&election), self.run_node(&election));
}
async fn run_election(&self, election: &Notify) {
loop {
tokio::select! {
biased;
_ = election.notified() => {}
_ = sleep(DNS_SERVER_ELECTION_INTERVAL) => {}
}
let mut rpc =
StandAloneServer::new(TcpTunnelListener::new(DNS_SERVER_RPC_ADDR.clone()));
if rpc.serve().await.is_err() {
// Another instance already owns the address — that's fine.
continue;
}
tracing::info!("won DNS server election, starting DnsServer");
let server = Arc::new(DnsServer::new(self.peer_mgr.clone(), rpc));
self.peer_mgr
.add_nic_packet_process_pipeline(Box::new(server.clone()))
.await;
server.run().await;
let _ = self
.peer_mgr
.remove_nic_packet_process_pipeline(server.id())
.await;
tracing::warn!("DnsServer exited, will retry election");
}
}
async fn run_node(&self, election: &Notify) {
let mut rpc = StandAloneClient::new(TcpTunnelConnector::new(DNS_SERVER_RPC_ADDR.clone()));
let mut heartbeat = HeartbeatRequest {
id: Some(self.id().into()),
@@ -55,7 +95,7 @@ impl DnsNode {
let sleep = sleep_until(last_heartbeat);
tokio::pin!(sleep);
let mut subscriber = self.mgr.get_global_ctx_ref().subscribe();
let mut subscriber = self.peer_mgr.get_global_ctx_ref().subscribe();
let mut tasks = JoinSet::new();
loop {
@@ -73,7 +113,7 @@ impl DnsNode {
_ = &mut sleep => {
if let Err(e) = self.heartbeat(&mut rpc, &mut heartbeat).await {
tracing::error!("heartbeat failed: {:?}", e);
self.election.notify_one();
election.notify_one();
}
last_heartbeat = Instant::now();
@@ -143,13 +183,13 @@ impl DnsNode {
}
fn refresh(&self, tasks: &mut JoinSet<()>, peer_ids: Vec<PeerId>) {
let my_peer_id = self.mgr.my_peer_id();
let my_peer_id = self.peer_mgr.my_peer_id();
for peer_id in peer_ids {
if peer_id == my_peer_id {
continue;
}
let mgr = self.mgr.clone();
let route = mgr.get_route();
let route = self.peer_mgr.get_route();
tasks.spawn(async move {
if let Some(peer_info) = route.get_peer_info(peer_id).await {
mgr.refresh(peer_id, peer_info.dns).await;