add more debug info for route (#155)

1. use info log for sync_route_info
2. allow dump route info with cli tool.
3. dump route info every 60s
This commit is contained in:
Sijie.Sun
2024-07-07 15:40:46 +08:00
committed by GitHub
parent 7cfa850d4c
commit 513e4cacc9
6 changed files with 178 additions and 110 deletions
+14 -17
View File
@@ -39,9 +39,7 @@ message PeerInfo {
message ListPeerRequest {} message ListPeerRequest {}
message ListPeerResponse { message ListPeerResponse { repeated PeerInfo peer_infos = 1; }
repeated PeerInfo peer_infos = 1;
}
enum NatType { enum NatType {
// has NAT; but own a single public IP, port is not changed // has NAT; but own a single public IP, port is not changed
@@ -77,13 +75,16 @@ message Route {
message ListRouteRequest {} message ListRouteRequest {}
message ListRouteResponse { message ListRouteResponse { repeated Route routes = 1; }
repeated Route routes = 1;
} message DumpRouteRequest {}
message DumpRouteResponse { string result = 1; }
service PeerManageRpc { service PeerManageRpc {
rpc ListPeer(ListPeerRequest) returns (ListPeerResponse); rpc ListPeer(ListPeerRequest) returns (ListPeerResponse);
rpc ListRoute(ListRouteRequest) returns (ListRouteResponse); rpc ListRoute(ListRouteRequest) returns (ListRouteResponse);
rpc DumpRoute(DumpRouteRequest) returns (DumpRouteResponse);
} }
enum ConnectorStatus { enum ConnectorStatus {
@@ -99,9 +100,7 @@ message Connector {
message ListConnectorRequest {} message ListConnectorRequest {}
message ListConnectorResponse { message ListConnectorResponse { repeated Connector connectors = 1; }
repeated Connector connectors = 1;
}
enum ConnectorManageAction { enum ConnectorManageAction {
ADD = 0; ADD = 0;
@@ -120,9 +119,7 @@ service ConnectorManageRpc {
rpc ManageConnector(ManageConnectorRequest) returns (ManageConnectorResponse); rpc ManageConnector(ManageConnectorRequest) returns (ManageConnectorResponse);
} }
message DirectConnectedPeerInfo { message DirectConnectedPeerInfo { int32 latency_ms = 1; }
int32 latency_ms = 1;
}
message PeerInfoForGlobalMap { message PeerInfoForGlobalMap {
map<uint32, DirectConnectedPeerInfo> direct_peers = 1; map<uint32, DirectConnectedPeerInfo> direct_peers = 1;
@@ -135,7 +132,8 @@ message GetGlobalPeerMapResponse {
} }
service PeerCenterRpc { service PeerCenterRpc {
rpc GetGlobalPeerMap (GetGlobalPeerMapRequest) returns (GetGlobalPeerMapResponse); rpc GetGlobalPeerMap(GetGlobalPeerMapRequest)
returns (GetGlobalPeerMapResponse);
} }
message VpnPortalInfo { message VpnPortalInfo {
@@ -145,12 +143,11 @@ message VpnPortalInfo {
} }
message GetVpnPortalInfoRequest {} message GetVpnPortalInfoRequest {}
message GetVpnPortalInfoResponse { message GetVpnPortalInfoResponse { VpnPortalInfo vpn_portal_info = 1; }
VpnPortalInfo vpn_portal_info = 1;
}
service VpnPortalRpc { service VpnPortalRpc {
rpc GetVpnPortalInfo (GetVpnPortalInfoRequest) returns (GetVpnPortalInfoResponse); rpc GetVpnPortalInfo(GetVpnPortalInfoRequest)
returns (GetVpnPortalInfoResponse);
} }
message HandshakeRequest { message HandshakeRequest {
+25 -4
View File
@@ -42,7 +42,7 @@ enum SubCommand {
Peer(PeerArgs), Peer(PeerArgs),
Connector(ConnectorArgs), Connector(ConnectorArgs),
Stun, Stun,
Route, Route(RouteArgs),
PeerCenter, PeerCenter,
VpnPortal, VpnPortal,
} }
@@ -72,6 +72,18 @@ enum PeerSubCommand {
List(PeerListArgs), List(PeerListArgs),
} }
#[derive(Args, Debug)]
struct RouteArgs {
#[command(subcommand)]
sub_command: Option<RouteSubCommand>,
}
#[derive(Subcommand, Debug)]
enum RouteSubCommand {
List,
Dump,
}
#[derive(Args, Debug)] #[derive(Args, Debug)]
struct ConnectorArgs { struct ConnectorArgs {
#[arg(short, long)] #[arg(short, long)]
@@ -204,6 +216,14 @@ impl CommandHandler {
Ok(()) Ok(())
} }
async fn handle_route_dump(&self) -> Result<(), Error> {
let mut client = self.get_peer_manager_client().await?;
let request = tonic::Request::new(DumpRouteRequest::default());
let response = client.dump_route(request).await?;
println!("response: {}", response.into_inner().result);
Ok(())
}
async fn handle_route_list(&self) -> Result<(), Error> { async fn handle_route_list(&self) -> Result<(), Error> {
#[derive(tabled::Tabled)] #[derive(tabled::Tabled)]
struct RouteTableItem { struct RouteTableItem {
@@ -307,9 +327,10 @@ async fn main() -> Result<(), Error> {
handler.handle_connector_list().await?; handler.handle_connector_list().await?;
} }
}, },
SubCommand::Route => { SubCommand::Route(route_args) => match route_args.sub_command {
handler.handle_route_list().await?; Some(RouteSubCommand::List) | None => handler.handle_route_list().await?,
} Some(RouteSubCommand::Dump) => handler.handle_route_dump().await?,
},
SubCommand::Stun => { SubCommand::Stun => {
timeout(Duration::from_secs(5), async move { timeout(Duration::from_secs(5), async move {
let collector = StunInfoCollector::new_with_default_servers(); let collector = StunInfoCollector::new_with_default_servers();
+4
View File
@@ -535,6 +535,10 @@ impl PeerManager {
self.get_route().list_routes().await self.get_route().list_routes().await
} }
pub async fn dump_route(&self) -> String {
self.get_route().dump().await
}
async fn run_nic_packet_process_pipeline(&self, data: &mut ZCPacket) { async fn run_nic_packet_process_pipeline(&self, data: &mut ZCPacket) {
for pipeline in self.nic_packet_process_pipeline.read().await.iter().rev() { for pipeline in self.nic_packet_process_pipeline.read().await.iter().rev() {
pipeline.try_process_packet_from_nic(data).await; pipeline.try_process_packet_from_nic(data).await;
+45 -11
View File
@@ -700,9 +700,13 @@ impl Debug for PeerRouteServiceImpl {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("PeerRouteServiceImpl") f.debug_struct("PeerRouteServiceImpl")
.field("my_peer_id", &self.my_peer_id) .field("my_peer_id", &self.my_peer_id)
.field("sessions", &self.sessions)
.field("route_table", &self.route_table)
.field("route_table_with_cost", &self.route_table_with_cost)
.field("synced_route_info", &self.synced_route_info)
.field( .field(
"sessions", "cached_local_conn_map",
&self.sessions.iter().map(|x| *x.key()).collect::<Vec<_>>(), &self.cached_local_conn_map.lock().unwrap(),
) )
.finish() .finish()
} }
@@ -842,9 +846,7 @@ impl PeerRouteServiceImpl {
let all_peer_ids = &conn_bitmap.peer_ids; let all_peer_ids = &conn_bitmap.peer_ids;
for (peer_idx, (peer_id, _)) in all_peer_ids.iter().enumerate() { for (peer_idx, (peer_id, _)) in all_peer_ids.iter().enumerate() {
let Some(connected) = self.synced_route_info.conn_map.get(peer_id) else { let connected = self.synced_route_info.conn_map.get(peer_id).unwrap();
continue;
};
for (idx, (other_peer_id, _)) in all_peer_ids.iter().enumerate() { for (idx, (other_peer_id, _)) in all_peer_ids.iter().enumerate() {
if connected.0.contains(other_peer_id) { if connected.0.contains(other_peer_id) {
@@ -942,7 +944,7 @@ impl PeerRouteServiceImpl {
let my_peer_id = self.my_peer_id; let my_peer_id = self.my_peer_id;
let (peer_infos, conn_bitmap) = self.build_sync_request(&session); let (peer_infos, conn_bitmap) = self.build_sync_request(&session);
tracing::trace!("my_id {:?}, pper_id: {:?}, peer_infos: {:?}, conn_bitmap: {:?}, synced_route_info: {:?} session: {:?}", tracing::info!("my_id {:?}, pper_id: {:?}, peer_infos: {:?}, conn_bitmap: {:?}, synced_route_info: {:?} session: {:?}",
my_peer_id, dst_peer_id, peer_infos, conn_bitmap, self.synced_route_info, session); my_peer_id, dst_peer_id, peer_infos, conn_bitmap, self.synced_route_info, session);
if peer_infos.is_none() if peer_infos.is_none()
@@ -952,6 +954,10 @@ impl PeerRouteServiceImpl {
return true; return true;
} }
session
.need_sync_initiator_info
.store(false, Ordering::Relaxed);
let ret = peer_rpc let ret = peer_rpc
.do_client_rpc_scoped(SERVICE_ID, dst_peer_id, |c| async { .do_client_rpc_scoped(SERVICE_ID, dst_peer_id, |c| async {
let client = RouteServiceClient::new(tarpc::client::Config::default(), c).spawn(); let client = RouteServiceClient::new(tarpc::client::Config::default(), c).spawn();
@@ -978,10 +984,6 @@ impl PeerRouteServiceImpl {
.dst_is_initiator .dst_is_initiator
.store(ret.is_initiator, Ordering::Relaxed); .store(ret.is_initiator, Ordering::Relaxed);
session
.need_sync_initiator_info
.store(false, Ordering::Relaxed);
session.update_dst_session_id(ret.session_id); session.update_dst_session_id(ret.session_id);
if let Some(peer_infos) = &peer_infos { if let Some(peer_infos) = &peer_infos {
@@ -1014,6 +1016,22 @@ struct RouteSessionManager {
sync_now_broadcast: tokio::sync::broadcast::Sender<()>, sync_now_broadcast: tokio::sync::broadcast::Sender<()>,
} }
impl Debug for RouteSessionManager {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("RouteSessionManager")
.field(
"session_tasks",
&self
.session_tasks
.iter()
.map(|x| *x.key())
.collect::<Vec<_>>(),
)
.field("dump_sessions", &self.dump_sessions())
.finish()
}
}
#[tarpc::server] #[tarpc::server]
impl RouteService for RouteSessionManager { impl RouteService for RouteSessionManager {
async fn sync_route_info( async fn sync_route_info(
@@ -1052,7 +1070,7 @@ impl RouteService for RouteSessionManager {
service_impl.update_route_table_and_cached_local_conn_bitmap(); service_impl.update_route_table_and_cached_local_conn_bitmap();
tracing::debug!( tracing::info!(
"sync_route_info: from_peer_id: {:?}, is_initiator: {:?}, peer_infos: {:?}, conn_bitmap: {:?}, synced_route_info: {:?} session: {:?}, new_route_table: {:?}", "sync_route_info: from_peer_id: {:?}, is_initiator: {:?}, peer_infos: {:?}, conn_bitmap: {:?}, synced_route_info: {:?} session: {:?}, new_route_table: {:?}",
from_peer_id, is_initiator, peer_infos, conn_bitmap, service_impl.synced_route_info, session, service_impl.route_table); from_peer_id, is_initiator, peer_infos, conn_bitmap, service_impl.synced_route_info, session, service_impl.route_table);
@@ -1297,6 +1315,16 @@ pub struct PeerRoute {
tasks: std::sync::Mutex<JoinSet<()>>, tasks: std::sync::Mutex<JoinSet<()>>,
} }
impl Debug for PeerRoute {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("PeerRoute")
.field("my_peer_id", &self.my_peer_id)
.field("service_impl", &self.service_impl)
.field("session_mgr", &self.session_mgr)
.finish()
}
}
impl PeerRoute { impl PeerRoute {
pub fn new( pub fn new(
my_peer_id: PeerId, my_peer_id: PeerId,
@@ -1322,6 +1350,8 @@ impl PeerRoute {
loop { loop {
tokio::time::sleep(Duration::from_secs(60)).await; tokio::time::sleep(Duration::from_secs(60)).await;
service_impl.clear_expired_peer(); service_impl.clear_expired_peer();
// TODO: use debug log level for this.
tracing::info!(?service_impl, "clear_expired_peer");
} }
} }
@@ -1449,6 +1479,10 @@ impl Route for PeerRoute {
*self.service_impl.cost_calculator.lock().unwrap() = Some(_cost_fn); *self.service_impl.cost_calculator.lock().unwrap() = Some(_cost_fn);
self.service_impl.update_route_table(); self.service_impl.update_route_table();
} }
async fn dump(&self) -> String {
format!("{:#?}", self)
}
} }
impl PeerPacketFilter for Arc<PeerRoute> {} impl PeerPacketFilter for Arc<PeerRoute> {}
+4
View File
@@ -78,6 +78,10 @@ pub trait Route {
} }
async fn set_route_cost_fn(&self, _cost_fn: RouteCostCalculator) {} async fn set_route_cost_fn(&self, _cost_fn: RouteCostCalculator) {}
async fn dump(&self) -> String {
"this route implementation does not support dump".to_string()
}
} }
pub type ArcRoute = Arc<Box<dyn Route + Send + Sync>>; pub type ArcRoute = Arc<Box<dyn Route + Send + Sync>>;
+11 -3
View File
@@ -1,9 +1,8 @@
use std::sync::Arc; use std::sync::Arc;
use crate::rpc::{ use crate::rpc::{
cli::PeerInfo, cli::PeerInfo, peer_manage_rpc_server::PeerManageRpc, DumpRouteRequest, DumpRouteResponse,
peer_manage_rpc_server::PeerManageRpc, ListPeerRequest, ListPeerResponse, ListRouteRequest, ListRouteResponse,
{ListPeerRequest, ListPeerResponse, ListRouteRequest, ListRouteResponse},
}; };
use tonic::{Request, Response, Status}; use tonic::{Request, Response, Status};
@@ -60,4 +59,13 @@ impl PeerManageRpc for PeerManagerRpcService {
reply.routes = self.peer_manager.list_routes().await; reply.routes = self.peer_manager.list_routes().await;
Ok(Response::new(reply)) Ok(Response::new(reply))
} }
async fn dump_route(
&self,
_request: Request<DumpRouteRequest>, // Accept request of type HelloRequest
) -> Result<Response<DumpRouteResponse>, Status> {
let mut reply = DumpRouteResponse::default();
reply.result = self.peer_manager.dump_route().await;
Ok(Response::new(reply))
}
} }