rename chan

in some cases iOS and macOS may be locked when looking for management domains during network changes

This change introduce an additional timeout on top of the context call

client/internal/debug/debug.go

@@ -56,6 +56,7 @@ block.prof: Block profiling information.
 heap.prof: Heap profiling information (snapshot of memory allocations).
 allocs.prof: Allocations profiling information.
 threadcreate.prof: Thread creation profiling information.
+stack_trace.txt: Complete stack traces of all goroutines at the time of bundle creation.
 
 
 Anonymization Process
@@ -109,6 +110,9 @@ go tool pprof -http=:8088 heap.prof
 
 This will open a web browser tab with the profiling information.
 
+Stack Trace
+The stack_trace.txt file contains a complete snapshot of all goroutine stack traces at the time the debug bundle was created.
+
 Routes
 The routes.txt file contains detailed routing table information in a tabular format:
 
@@ -327,6 +331,10 @@ func (g *BundleGenerator) createArchive() error {
 		log.Errorf("failed to add profiles to debug bundle: %v", err)
 	}
 
+	if err := g.addStackTrace(); err != nil {
+		log.Errorf("failed to add stack trace to debug bundle: %v", err)
+	}
+
 	if err := g.addSyncResponse(); err != nil {
 		return fmt.Errorf("add sync response: %w", err)
 	}
@@ -522,6 +530,18 @@ func (g *BundleGenerator) addProf() (err error) {
 	return nil
 }
 
+func (g *BundleGenerator) addStackTrace() error {
+	buf := make([]byte, 5242880) // 5 MB buffer
+	n := runtime.Stack(buf, true)
+
+	stackTrace := bytes.NewReader(buf[:n])
+	if err := g.addFileToZip(stackTrace, "stack_trace.txt"); err != nil {
+		return fmt.Errorf("add stack trace file to zip: %w", err)
+	}
+
+	return nil
+}
+
 func (g *BundleGenerator) addInterfaces() error {
 	interfaces, err := net.Interfaces()
 	if err != nil {

client/internal/dns/mgmt/mgmt.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"net"
+	"net/netip"
 	"net/url"
 	"strings"
 	"sync"
@@ -26,6 +27,11 @@ type Resolver struct {
 	mutex         sync.RWMutex
 }
 
+type ipsResponse struct {
+	ips []netip.Addr
+	err error
+}
+
 // NewResolver creates a new management domains cache resolver.
 func NewResolver() *Resolver {
 	return &Resolver{
@@ -99,9 +105,9 @@ func (m *Resolver) AddDomain(ctx context.Context, d domain.Domain) error {
 	ctx, cancel := context.WithTimeout(ctx, dnsTimeout)
 	defer cancel()
 
-	ips, err := net.DefaultResolver.LookupNetIP(ctx, "ip", d.PunycodeString())
+	ips, err := lookupIPWithExtraTimeout(ctx, d)
 	if err != nil {
-		return fmt.Errorf("resolve domain %s: %w", d.SafeString(), err)
+		return err
 	}
 
 	var aRecords, aaaaRecords []dns.RR
@@ -159,6 +165,36 @@ func (m *Resolver) AddDomain(ctx context.Context, d domain.Domain) error {
 	return nil
 }
 
+func lookupIPWithExtraTimeout(ctx context.Context, d domain.Domain) ([]netip.Addr, error) {
+	log.Infof("looking up IP for mgmt domain=%s", d.SafeString())
+	defer log.Infof("done looking up IP for mgmt domain=%s", d.SafeString())
+	resultChan := make(chan *ipsResponse, 1)
+
+	go func() {
+		ips, err := net.DefaultResolver.LookupNetIP(ctx, "ip", d.PunycodeString())
+		resultChan <- &ipsResponse{
+			err: err,
+			ips: ips,
+		}
+	}()
+
+	var resp *ipsResponse
+
+	select {
+	case <-time.After(dnsTimeout + time.Millisecond*500):
+		log.Warnf("timed out waiting for IP for mgmt domain=%s", d.SafeString())
+		return nil, fmt.Errorf("timed out waiting for ips to be available for domain %s", d.SafeString())
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	case resp = <-resultChan:
+	}
+
+	if resp.err != nil {
+		return nil, fmt.Errorf("resolve domain %s: %w", d.SafeString(), resp.err)
+	}
+	return resp.ips, nil
+}
+
 // PopulateFromConfig extracts and caches domains from the client configuration.
 func (m *Resolver) PopulateFromConfig(ctx context.Context, mgmtURL *url.URL) error {
 	if mgmtURL == nil {

client/internal/peer/endpoint.go

@@ -20,7 +20,7 @@ type EndpointUpdater struct {
 	wgConfig  WgConfig
 	initiator bool
 
-	// mu protects updateWireGuardPeer and cancelFunc
+	// mu protects cancelFunc
 	mu         sync.Mutex
 	cancelFunc func()
 	updateWg   sync.WaitGroup
@@ -86,11 +86,9 @@ func (e *EndpointUpdater) scheduleDelayedUpdate(ctx context.Context, addr *net.U
 	case <-ctx.Done():
 		return
 	case <-t.C:
-		e.mu.Lock()
 		if err := e.updateWireGuardPeer(addr, presharedKey); err != nil {
 			e.log.Errorf("failed to update WireGuard peer, address: %s, error: %v", addr, err)
 		}
-		e.mu.Unlock()
 	}
 }
 

formatter/hook/hook.go

@@ -60,14 +60,7 @@ func (hook ContextHook) Fire(entry *logrus.Entry) error {
 
 	entry.Data["context"] = source
 
-	switch source {
-	case HTTPSource:
-		addHTTPFields(entry)
-	case GRPCSource:
-		addGRPCFields(entry)
-	case SystemSource:
-		addSystemFields(entry)
-	}
+	addFields(entry)
 
 	return nil
 }
@@ -99,7 +92,7 @@ func (hook ContextHook) parseSrc(filePath string) string {
 	return fmt.Sprintf("%s/%s", pkg, file)
 }
 
-func addHTTPFields(entry *logrus.Entry) {
+func addFields(entry *logrus.Entry) {
 	if ctxReqID, ok := entry.Context.Value(context.RequestIDKey).(string); ok {
 		entry.Data[context.RequestIDKey] = ctxReqID
 	}
@@ -109,30 +102,6 @@ func addHTTPFields(entry *logrus.Entry) {
 	if ctxInitiatorID, ok := entry.Context.Value(context.UserIDKey).(string); ok {
 		entry.Data[context.UserIDKey] = ctxInitiatorID
 	}
-}
-
-func addGRPCFields(entry *logrus.Entry) {
-	if ctxReqID, ok := entry.Context.Value(context.RequestIDKey).(string); ok {
-		entry.Data[context.RequestIDKey] = ctxReqID
-	}
-	if ctxAccountID, ok := entry.Context.Value(context.AccountIDKey).(string); ok {
-		entry.Data[context.AccountIDKey] = ctxAccountID
-	}
-	if ctxDeviceID, ok := entry.Context.Value(context.PeerIDKey).(string); ok {
-		entry.Data[context.PeerIDKey] = ctxDeviceID
-	}
-}
-
-func addSystemFields(entry *logrus.Entry) {
-	if ctxReqID, ok := entry.Context.Value(context.RequestIDKey).(string); ok {
-		entry.Data[context.RequestIDKey] = ctxReqID
-	}
-	if ctxInitiatorID, ok := entry.Context.Value(context.UserIDKey).(string); ok {
-		entry.Data[context.UserIDKey] = ctxInitiatorID
-	}
-	if ctxAccountID, ok := entry.Context.Value(context.AccountIDKey).(string); ok {
-		entry.Data[context.AccountIDKey] = ctxAccountID
-	}
 	if ctxDeviceID, ok := entry.Context.Value(context.PeerIDKey).(string); ok {
 		entry.Data[context.PeerIDKey] = ctxDeviceID
 	}

management/internals/server/boot.go

@@ -10,9 +10,9 @@ import (
 	"slices"
 	"time"
 
-	"github.com/google/uuid"
 	grpcMiddleware "github.com/grpc-ecosystem/go-grpc-middleware/v2"
 	"github.com/grpc-ecosystem/go-grpc-middleware/v2/interceptors/realip"
+	"github.com/rs/xid"
 	log "github.com/sirupsen/logrus"
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/credentials"
@@ -180,7 +180,7 @@ func unaryInterceptor(
 	info *grpc.UnaryServerInfo,
 	handler grpc.UnaryHandler,
 ) (interface{}, error) {
-	reqID := uuid.New().String()
+	reqID := xid.New().String()
 	//nolint
 	ctx = context.WithValue(ctx, hook.ExecutionContextKey, hook.GRPCSource)
 	//nolint
@@ -194,7 +194,7 @@ func streamInterceptor(
 	info *grpc.StreamServerInfo,
 	handler grpc.StreamHandler,
 ) error {
-	reqID := uuid.New().String()
+	reqID := xid.New().String()
 	wrapped := grpcMiddleware.WrapServerStream(ss)
 	//nolint
 	ctx := context.WithValue(ss.Context(), hook.ExecutionContextKey, hook.GRPCSource)

management/server/telemetry/http_api_metrics.go

@@ -7,8 +7,8 @@ import (
 	"strings"
 	"time"
 
-	"github.com/google/uuid"
 	"github.com/gorilla/mux"
+	"github.com/rs/xid"
 	log "github.com/sirupsen/logrus"
 	"go.opentelemetry.io/otel/attribute"
 	"go.opentelemetry.io/otel/metric"
@@ -169,7 +169,7 @@ func (m *HTTPMiddleware) Handler(h http.Handler) http.Handler {
 		//nolint
 		ctx := context.WithValue(r.Context(), hook.ExecutionContextKey, hook.HTTPSource)
 
-		reqID := uuid.New().String()
+		reqID := xid.New().String()
 		//nolint
 		ctx = context.WithValue(ctx, nbContext.RequestIDKey, reqID)