Reduce relay log spam (#1733)

relay_manager.go

@@ -7,6 +7,7 @@ import (
 	"fmt"
 	"log/slog"
 	"net/netip"
+	"slices"
 	"sync/atomic"
 
 	"github.com/slackhq/nebula/cert"
@@ -57,14 +58,25 @@ func (rm *relayManager) GetUseRelays() bool {
 // For each candidate relay it either kicks off a handshake to the relay, sends a CreateRelayRequest, retransmits
 // one that may have been lost, or, once the relay is Established, forwards the in-progress
 // stage 0 handshake packet for vpnIp through it.
-func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hostinfo *HostInfo, stage0 []byte) {
+func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hh *HandshakeHostInfo, stage0 []byte) {
+	hostinfo := hh.hostinfo
 	if !rm.GetUseRelays() || len(hostinfo.remotes.relays) == 0 {
+		hh.lastRelays = nil
 		return
 	}
 
-	hostinfo.logger(rm.l).Info("Attempt to relay through hosts", "relays", hostinfo.remotes.relays)
+	relays := hostinfo.remotes.relays
+	listLevel := slog.LevelDebug
+	prior := hh.lastRelays
+	if !slices.Equal(relays, prior) {
+		listLevel = slog.LevelInfo
+		hh.lastRelays = slices.Clone(relays)
+	}
+	hl := hostinfo.logger(rm.l)
+	hl.Log(context.Background(), listLevel, "Attempt to relay through hosts", "relays", relays)
+
 	// Send a RelayRequest to all known Relay IP's
-	for _, relay := range hostinfo.remotes.relays {
+	for _, relay := range relays {
 		// Don't relay through the host I'm trying to connect to
 		if relay == vpnIp {
 			continue
@@ -75,12 +87,19 @@ func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hostinfo *Ho
 			continue
 		}
 
+		// Each relay's per-attempt log fires at Info on the first time we hit it and Debug after that.
+		level := slog.LevelInfo
+		if slices.Contains(prior, relay) {
+			level = slog.LevelDebug
+		}
+
 		relayHostInfo := rm.hostmap.QueryVpnAddr(relay)
 		if relayHostInfo == nil || !relayHostInfo.remote.IsValid() {
-			hostinfo.logger(rm.l).Info("Establish tunnel to relay target", "relay", relay.String())
+			hl.Log(context.Background(), level, "Establish tunnel to relay target", "relay", relay.String())
 			f.Handshake(relay)
 			continue
 		}
+
 		// Check the relay HostInfo to see if we already established a relay through
 		existingRelay, ok := relayHostInfo.relayState.QueryRelayForByIp(vpnIp)
 		if !ok {
@@ -88,7 +107,7 @@ func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hostinfo *Ho
 			if relayHostInfo.remote.IsValid() {
 				idx, err := AddRelay(rm.l, relayHostInfo, rm.hostmap, vpnIp, nil, TerminalType, Requested)
 				if err != nil {
-					hostinfo.logger(rm.l).Info("Failed to add relay to hostmap", "relay", relay.String(), "error", err)
+					hl.Info("Failed to add relay to hostmap", "relay", relay.String(), "error", err)
 				}
 
 				m := NebulaControl{
@@ -99,12 +118,12 @@ func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hostinfo *Ho
 				switch relayHostInfo.GetCert().Certificate.Version() {
 				case cert.Version1:
 					if !f.myVpnAddrs[0].Is4() {
-						hostinfo.logger(rm.l).Error("can not establish v1 relay with a v6 network because the relay is not running a current nebula version")
+						hl.Error("can not establish v1 relay with a v6 network because the relay is not running a current nebula version")
 						continue
 					}
 
 					if !vpnIp.Is4() {
-						hostinfo.logger(rm.l).Error("can not establish v1 relay with a v6 remote network because the relay is not running a current nebula version")
+						hl.Error("can not establish v1 relay with a v6 remote network because the relay is not running a current nebula version")
 						continue
 					}
 
@@ -116,16 +135,16 @@ func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hostinfo *Ho
 					m.RelayFromAddr = netAddrToProtoAddr(f.myVpnAddrs[0])
 					m.RelayToAddr = netAddrToProtoAddr(vpnIp)
 				default:
-					hostinfo.logger(rm.l).Error("Unknown certificate version found while creating relay")
+					hl.Error("Unknown certificate version found while creating relay")
 					continue
 				}
 
 				msg, err := m.Marshal()
 				if err != nil {
-					hostinfo.logger(rm.l).Error("Failed to marshal Control message to create relay", "error", err)
+					hl.Error("Failed to marshal Control message to create relay", "error", err)
 				} else {
 					f.SendMessageToHostInfo(header.Control, 0, relayHostInfo, msg, make([]byte, 12), make([]byte, mtu))
-					rm.l.Info("send CreateRelayRequest",
+					rm.l.Log(context.Background(), level, "send CreateRelayRequest",
 						"relayFrom", f.myVpnAddrs[0],
 						"relayTo", vpnIp,
 						"initiatorRelayIndex", idx,
@@ -138,14 +157,14 @@ func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hostinfo *Ho
 
 		switch existingRelay.State {
 		case Established:
-			hostinfo.logger(rm.l).Info("Send handshake via relay", "relay", relay.String())
+			hl.Log(context.Background(), level, "Send handshake via relay", "relay", relay.String())
 			f.SendVia(relayHostInfo, existingRelay, stage0, make([]byte, 12), make([]byte, mtu), false)
 		case Disestablished:
 			// Mark this relay as 'requested'
 			relayHostInfo.relayState.UpdateRelayForByIpState(vpnIp, Requested)
 			fallthrough
 		case Requested:
-			hostinfo.logger(rm.l).Info("Re-send CreateRelay request", "relay", relay.String())
+			hl.Log(context.Background(), level, "Re-send CreateRelay request", "relay", relay.String())
 			// Re-send the CreateRelay request, in case the previous one was lost.
 			m := NebulaControl{
 				Type:                NebulaControl_CreateRelayRequest,
@@ -155,12 +174,12 @@ func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hostinfo *Ho
 			switch relayHostInfo.GetCert().Certificate.Version() {
 			case cert.Version1:
 				if !f.myVpnAddrs[0].Is4() {
-					hostinfo.logger(rm.l).Error("can not establish v1 relay with a v6 network because the relay is not running a current nebula version")
+					hl.Error("can not establish v1 relay with a v6 network because the relay is not running a current nebula version")
 					continue
 				}
 
 				if !vpnIp.Is4() {
-					hostinfo.logger(rm.l).Error("can not establish v1 relay with a v6 remote network because the relay is not running a current nebula version")
+					hl.Error("can not establish v1 relay with a v6 remote network because the relay is not running a current nebula version")
 					continue
 				}
 
@@ -172,16 +191,16 @@ func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hostinfo *Ho
 				m.RelayFromAddr = netAddrToProtoAddr(f.myVpnAddrs[0])
 				m.RelayToAddr = netAddrToProtoAddr(vpnIp)
 			default:
-				hostinfo.logger(rm.l).Error("Unknown certificate version found while creating relay")
+				hl.Error("Unknown certificate version found while creating relay")
 				continue
 			}
 			msg, err := m.Marshal()
 			if err != nil {
-				hostinfo.logger(rm.l).Error("Failed to marshal Control message to create relay", "error", err)
+				hl.Error("Failed to marshal Control message to create relay", "error", err)
 			} else {
 				// This must send over the hostinfo, not over hm.Hosts[ip]
 				f.SendMessageToHostInfo(header.Control, 0, relayHostInfo, msg, make([]byte, 12), make([]byte, mtu))
-				rm.l.Info("send CreateRelayRequest",
+				rm.l.Log(context.Background(), level, "send CreateRelayRequest",
 					"relayFrom", f.myVpnAddrs[0],
 					"relayTo", vpnIp,
 					"initiatorRelayIndex", existingRelay.LocalIndex,
@@ -192,7 +211,7 @@ func (rm *relayManager) StartRelays(f *Interface, vpnIp netip.Addr, hostinfo *Ho
 			// PeerRequested only occurs in Forwarding relays, not Terminal relays, and this is a Terminal relay case.
 			fallthrough
 		default:
-			hostinfo.logger(rm.l).Error("Relay unexpected state",
+			hl.Error("Relay unexpected state",
 				"vpnIp", vpnIp,
 				"state", existingRelay.State,
 				"relay", relay,