MDC log link design


The existing log system in our project adopts the set of log components slf4j+logback, which is also a commonly used log component in Java ecology. However, with the evolution of distribution, this set of components obviously has the following problems:

1. Various irrelevant logs pass through them, which may make it impossible for us to directly locate the whole operation process. Therefore, we may need to classify and mark a user's operation process, that is, add a unique ID on its log information, such as using thread + timestamp, or user ID; grep the operation flow of a user from a large amount of log information.
2. It is not possible to do information embedding, so it is not convenient to do follow-up system and business analysis
3. Log troubleshooting is inconvenient. You need to export or view logs online through linux commands


   When the author was in Ctrip group, a large number of middleware had been incubated internally. Among them, the distributed log component had been applied to different applications under major business departments. According to statistics, tens of thousands of applications in the whole group were connected to this log component, and a design diagram was drawn according to the impression



The theme of this blog is MDC   Mapped Diagnostic Context can be roughly understood as a thread safe container for storing diagnostic logs). Its specific process is to string the whole track through some identifiers, such as A-B-C-remote interface-D. the link related log information can be quickly found in the log file through a certain identifier. The following describes the log scheme in the project I am currently responsible for


Configuring traceId in logback.xml is a bit like a placeholder



      Write the corresponding traceId variable through MDC


Source code analysis

1. What is MDC?

The following figure shows that MDC is a class of slf4j API, which provides put,get,remove and other methods. After reading the source code, we can see that it is actually a ThreadLocal. Each put element is put into it. When is called, take out the ThreadLocal variable and assign it to the output log






It can be seen from the above

1 MDCAdapter is an adaptation interface, which is stored under the spi package. Therefore, it is known that MDCAdapter is used to adapt other log components

2. The put method provided by MDC can put a K-V Key value pair into the container, and ensure that the Key is unique in the same thread, and the MDC values of different threads do not affect each other

3 in logback.xml, you can output the req in MDC by declaring% X{REQ_ID} in layout_ ID information

4. The remove method provided by MDC can clear the key value pair information corresponding to the specified key in MDC


LogbackMDCAdapters source code



The above is the use method and source code analysis of MDC. The following describes how to extract and package the logs before and after the call and print them uniformly when calling an external system locally. Because the author did not extract the previous code, each different calling method has to write manually, Although this method has no big problem, it is obviously redundant and can be extracted


External interface log trace output

The external interface is involved in the calling process. Because the external interface is in a third-party system, we cannot pass the traceId down. We need to modify our remote calling code. Because the author's project uses restTemplate, we need to add an interceptor to restTemplate to print out relevant logs before and after sending the request, The following is the log interceptor corresponding to my restTemplate

class MyRequestInterceptor implements ClientHttpRequestInterceptor {

        public ClientHttpResponse intercept(HttpRequest request, byte[] bytes, ClientHttpRequestExecution execution) throws IOException {
            traceRequest(request, bytes);

            ClientHttpResponse response = execution.execute(request, bytes);
            ClientHttpResponse responseCopy = new BufferingClientHttpResponseWrapper(response);
            return responseCopy;

         * Print request data
         * @param request request
         * @param bytes   Request body
        private void traceRequest(HttpRequest request, byte[] bytes) {
            String body = new String(bytes, StandardCharsets.UTF_8);
  "Request Body = {}", body);

         * Print response results
         * @param response Response results
         * @throws IOException io
        private void traceResponse(ClientHttpResponse response) throws IOException {
            StringBuilder inputStringBuilder = new StringBuilder();
            try (BufferedReader bufferedReader =
                         new BufferedReader(new InputStreamReader(response.getBody(), StandardCharsets.UTF_8))) {
                String line = bufferedReader.readLine();
                while (line != null) {
                    // inputStringBuilder.append('\n');
                    line = bufferedReader.readLine();
  "Response Body: {}", inputStringBuilder.toString());

        final class BufferingClientHttpResponseWrapper implements ClientHttpResponse {
            private final ClientHttpResponse response;
            private byte[] body;

            BufferingClientHttpResponseWrapper(ClientHttpResponse response) {
                this.response = response;

            public HttpStatus getStatusCode() throws IOException {
                return this.response.getStatusCode();

            public int getRawStatusCode() throws IOException {
                return this.response.getRawStatusCode();

            public String getStatusText() throws IOException {
                return this.response.getStatusText();

            public HttpHeaders getHeaders() {
                return this.response.getHeaders();

            public InputStream getBody() throws IOException {
                if (this.body == null) {
                    this.body = StreamUtils.copyToByteArray(this.response.getBody());
                return new ByteArrayInputStream(this.body);

            public void close() {




  The above is about the common usage scenarios of MDC, including the log component in Ctrip. In fact, it is also implemented through MDC internally, but it is adjusted according to the business. In general, in a distributed environment, it is best to output the log to Redis or ES, and then provide an interface to query the log. At present, many similar open-source frameworks integrate distributed link log printing + kanban



Posted on Wed, 01 Dec 2021 15:15:14 -0500 by Demonic