[#107] size check before proto buffer encoding for big exception meta…

…data

errors.go

@@ -95,6 +95,9 @@ func (span *span) addCauserCallStack(err error, eid int64) {
 		if !ok {
 			break
 		}
+		if !span.canAddErrorChain() {
+			break
+		}
 
 		e = c.Cause()
 		if t := span.findError(e); t == nil {

grpc.go

@@ -452,12 +452,19 @@ func (agentGrpc *agentGrpc) sendSqlUidMetadataWithRetry(sqlUid []byte, sql strin
 }
 
 func (agentGrpc *agentGrpc) sendExceptionMetadata(ctx context.Context, in *pb.PExceptionMetaData) error {
+	// size check before proto buffer encoding to ensure grpc stability
+	if in.XXX_Size() >= grpcWriteBufferSize {
+		err := status.Errorf(codes.ResourceExhausted, "gRPC message exceeds maximum size: %d", grpcWriteBufferSize)
+		Log("grpc").Warnf("skip exception metadata - %v", err)
+		return err
+	}
+
 	ctx, cancel := context.WithTimeout(ctx, agentGrpcTimeOut)
 	defer cancel()
 
 	_, err := agentGrpc.metaClient.RequestExceptionMetaData(ctx, in)
 	if err != nil {
-		Log("grpc").Errorf("send sql metadata - %v", err)
+		Log("grpc").Errorf("send exception metadata - %v", err)
 	}
 
 	return err

span.go

@@ -26,6 +26,7 @@ const (
 	defaultEventDepth     = 64
 	defaultEventSequence  = 5000
 	defaultEventChunkSize = 20
+	maxErrorChainEntry    = 10
 )
 
 var (
@@ -127,8 +128,9 @@ func (span *span) EndSpan() {
 
 	chunk := span.newEventChunk(true)
 	if chunk.enqueue() {
-		if len(span.errorChains) > 0 {
+		if span.errorChains != nil && len(span.errorChains) > 0 {
 			span.agent.enqueueExceptionMeta(span)
+			span.errorChains = nil
 		}
 	} else {
 		Log("span").Tracef("span channel - max capacity reached or closed")
@@ -443,6 +445,10 @@ func (span *span) config() *Config {
 	return span.agent.config
 }
 
+func (span *span) canAddErrorChain() bool {
+	return span.errorChains != nil && len(span.errorChains) < maxErrorChainEntry
+}
+
 type spanChunk struct {
 	span       *span
 	eventChunk []*spanEvent

span_event.go

@@ -101,7 +101,7 @@ func (se *spanEvent) SetError(e error, errorName ...string) {
 	se.errorString = e.Error()
 
 	cfg := se.config()
-	if cfg.errorTraceCallStack {
+	if cfg.errorTraceCallStack && se.parentSpan.canAddErrorChain() {
 		se.exceptionId = se.parentSpan.traceCallStack(e, cfg.errorCallStackDepth)
 		se.Annotations().AppendLong(AnnotationExceptionChainId, se.exceptionId)
 	}