java.net.SocketException:Connection reset 原因分析与故障重现

本贴最后更新于 1631 天前,其中的信息可能已经水流花落

前两天同事问了一个 java.net.SocketException: Connection reset 的问题,场景是在一个 Java 应用中使用 httpClient 请求远程一个 RestfulApi 接口,但返回报错 Connection reset。由于没有到现场查看日志,只能通过查看源码和资料进行故障重现,本文记录了问题的分析过程。

异常信息

[2019-10-24 14:55:34.390][HttpClientUtils.java:797]-doPost 发送 POST 请求出现异常:请求参数:{aaa:bbb,ccc:ddd},url:http://xxx.xxx.com/xxx/xxx, 时间秒:0.019 秒,Connection reset java.net.SocketException: Connection reset

在源码中定位异常信息

查看 jdk10 源码,发现 Connection reset 信息出现于方法 java.net.SocketInputStream#read(byte[], int, int, int) 。

  • 此方法通过间接调用 socketRead-->socketRead0-->NET_Read-->read 读取内核态缓存区数据。
  • 如果 NET_Read 函数(unix 系统)检测 Connection reset 或 Broken pipe,将抛出 sun/net/ConnectionResetException,相应地在 jvm 中会抛出异常 SocketException("Connection reset")。
  • 详情请见下方两段源码中文注释部分。
    /**
     * Reads into a byte array <i>b</i> at offset <i>off</i>,
     * <i>length</i> bytes of data.
     * @param b the buffer into which the data is read
     * @param off the start offset of the data
     * @param length the maximum number of bytes read
     * @return the actual number of bytes read, -1 is
     *          returned when the end of the stream is reached.
     * @exception IOException If an I/O error has occurred.
     */
    public int read(byte b[], int off, int length) throws IOException {
        return read(b, off, length, impl.getTimeout());
    }

    int read(byte b[], int off, int length, int timeout) throws IOException {
        int n;

        // EOF already encountered
        if (eof) {
            return -1;
        }

        // connection reset
        if (impl.isConnectionReset()) {
            throw new SocketException("Connection reset");
        }

        // bounds check
        if (length <= 0 || off < 0 || length > b.length - off) {
            if (length == 0) {
                return 0;
            }
            throw new ArrayIndexOutOfBoundsException("length == " + length
                    + " off == " + off + " buffer length == " + b.length);
        }

        boolean gotReset = false;

        // acquire file descriptor and do the read
        FileDescriptor fd = impl.acquireFD();
        try {
            // 间接调用socketRead0函数,从内核态缓存区读取数据
            n = socketRead(fd, b, off, length, timeout);
            if (n > 0) {
                return n;
            }
        } catch (ConnectionResetException rstExc) {
            // 如果socketRead0抛出ConnectionResetException,此处进行捕获。后续再重试一次读取操作。
            gotReset = true;
        } finally {
            impl.releaseFD();
        }

        /*
         * We receive a "connection reset" but there may be bytes still
         * buffered on the socket
         * 重试读取操作
         */
        if (gotReset) {
            impl.setConnectionResetPending();
            impl.acquireFD();
            try {
                n = socketRead(fd, b, off, length, timeout);
                if (n > 0) {
                    return n;
                }
            } catch (ConnectionResetException rstExc) {
            } finally {
                impl.releaseFD();
            }
        }

        /*
         * If we get here we are at EOF, the socket has been closed,
         * or the connection has been reset.
         */
        if (impl.isClosedOrPending()) {
            throw new SocketException("Socket closed");
        }
        if (impl.isConnectionResetPending()) {
            impl.setConnectionReset();
        }
        if (impl.isConnectionReset()) {
            // 最终抛出异常
            throw new SocketException("Connection reset");
        }
        eof = true;
        return -1;
    }
/*
 * Class:     java_net_SocketInputStream
 * Method:    socketRead0
 * Signature: (Ljava/io/FileDescriptor;[BIII)I
 */
JNIEXPORT jint JNICALL
Java_java_net_SocketInputStream_socketRead0(JNIEnv *env, jobject this,
                                            jobject fdObj, jbyteArray data,
                                            jint off, jint len, jint timeout)
{
    char BUF[MAX_BUFFER_LEN];
    char *bufP;
    jint fd, nread;

    if (IS_NULL(fdObj)) {
        JNU_ThrowByName(env, "java/net/SocketException",
                        "Socket closed");
        return -1;
    }
    fd = (*env)->GetIntField(env, fdObj, IO_fd_fdID);
    if (fd == -1) {
        JNU_ThrowByName(env, "java/net/SocketException", "Socket closed");
        return -1;
    }

    /*
     * If the read is greater than our stack allocated buffer then
     * we allocate from the heap (up to a limit)
     */
    if (len > MAX_BUFFER_LEN) {
        if (len > MAX_HEAP_BUFFER_LEN) {
            len = MAX_HEAP_BUFFER_LEN;
        }
        bufP = (char *)malloc((size_t)len);
        if (bufP == NULL) {
            bufP = BUF;
            len = MAX_BUFFER_LEN;
        }
    } else {
        bufP = BUF;
    }
    if (timeout) {
        nread = NET_ReadWithTimeout(env, fd, bufP, len, timeout);
        if ((*env)->ExceptionCheck(env)) {
            if (bufP != BUF) {
                free(bufP);
            }
            return nread;
        }
    } else {
        nread = NET_Read(fd, bufP, len);
    }

    if (nread <= 0) {
        if (nread < 0) {

            switch (errno) {
                case ECONNRESET:
                case EPIPE:
                    /* 抛出异常的情况对应为Connection reset和Broken pipe
                    * ECONNRESET      Connection reset (POSIX.1-2001).
                    * EPIPE           Broken pipe (POSIX.1-2001).
                    * @see http://man7.org/linux/man-pages/man3/errno.3.html
                    */
                    JNU_ThrowByName(env, "sun/net/ConnectionResetException",
                        "Connection reset");
                    break;

                case EBADF:
                    JNU_ThrowByName(env, "java/net/SocketException",
                        "Socket closed");
                    break;

                case EINTR:
                     JNU_ThrowByName(env, "java/io/InterruptedIOException",
                           "Operation interrupted");
                     break;
                default:
                    JNU_ThrowByNameWithMessageAndLastError
                        (env, "java/net/SocketException", "Read failed");
            }
        }
    } else {
        (*env)->SetByteArrayRegion(env, data, off, nread, (jbyte *)bufP);
    }

    if (bufP != BUF) {
        free(bufP);
    }
    return nread;
}

java.net.SoecketException:Connection reset 出现原因

根据 IBM 开发者社区博客 Understanding some common SocketExceptions in JAVA 描述,在 Java 中出现 Connnection reset 的原因为本地 socket 在接收或发送数据时收到远端系统带 RST 位的数据包。

  • This exception appears when the remote connection is unexpectedly and forcefully closed due to various reasons like application crash, system reboot, hard close of remote host. Kernel from the remote system sends out a packets with RST bit to the local system. The local socket on performing any SEND (could be a Keep-alive packet) or RECEIVE operations subsequently fail with this error. Certain combinations of linger settings can also result in packets with RST bit set.

RST 报文段出现的场景

根据《TCP/IP 详解.卷一:协议》,复位报文段(RST)可能在以下三种情况下出现:

  1. 到不存在的端口的连接请求
    • 当连接请求到达时,目的端口没有进程正在监听。即 TCP 连接建立时,客户端发送 SYN 报文段,服务端的 TCP 端软件回复 RST 报文段,此过程对应 Socket Api 的 connect 函数,抛出异常为:Connection refused
    • 出现阶段:建立连接时
  2. 异常终止一个连接
    • 正常终止一个连接的方式是一方发送 FIN。因为在所有排队数据都已经发送之后才发送 FIN,正常情况下没有任何数据丢失。
    • 存在一种异常终止连接的方式:一方发送一个复位报文段(RST)而不是 FIN 来中途释放一个连接。其行为表现为:一方丢弃任何待发数据并立即发送复位报文段,收到 RST 的一方不会再回 ACK 报文段,而是终止该连接,并通知应用层连接复位,抛出异常 Connection reset by peer。
    • Socket Api 通过“linger on close”选项(SO_LINGER)提供了异常终止连接的能力。
    • 详情可以继续参考 TCP option SO_LINGER (zero) - when it's required
    • 出现阶段:终止连接时
  3. 检测半开连接
    • 半开连接:如果一方已经关闭或异常终止连接而另一方却不知道,这样的 TCP 连接称为半开连接。即异常的一方没有发出 FIN 报文段来通知对方开始终止连接。
    • 现象: 只要不打算在半开连接上传输数据,仍处于连接状态的一方就不会检测另一方已经出现异常。这个异常会在处于连接状态一方进行 read 或 write 时抛出,错误信息为 Connection reset 或 Connetion reset by peer。
    • 原因:异常一方已经丢失了连接的所有信息,它不知道另一方发送数据报文段中提到的连接。TCP 的处理原理是接收方以复位(RST)作为回应。
    • 什么时候有半开连接?更多情况可见 Detection of Half-Open (Dropped) Connections
      • application crash 应用崩溃
      • system reboot 系统重启(导致应用崩溃或导致应用崩溃后重启)
      • hard close of remote host (远程主机的硬关闭(掉电等))
      • Router crash/reboot 两方中间某跳路由器崩溃或重启
    • 出现阶段:读取或写入数据时

重现故障的方案

根据 RST 报文段出现的三种场景,分三种情况对故障进行重现。

到不存在的端口的连接请求

重现步骤
  1. 准备一个客户端 httpClient 请求示例,用于发起远程 RestfulApi 接口调用。示例程序链接见参考资料中的 connection-reset-demo。
/**
 * java.net.SocketException : Connection reset 测试客户端
 */
@SpringBootApplication
@RestController
public class ApplicationCenter {

    private static final Logger LOGGER = LoggerFactory.getLogger(ApplicationCenter.class);

    @Autowired
    private RestTemplate restTemplate;

    public static void main(String[] args) {
        SpringApplication.run(ApplicationCenter.class, args);
    }

    @Bean
    public RestTemplate restTemplate(RestTemplateBuilder builder){
        return builder.build();
    }

    @RequestMapping(path = "/hello", produces = MediaType.APPLICATION_JSON_VALUE)
    public String serverHello(@RequestParam("key")String key) {
        HttpHeaders headers = new HttpHeaders();
        headers.setContentType(MediaType.APPLICATION_FORM_URLENCODED);
        MultiValueMap<String, String> map= new LinkedMultiValueMap<>();
        map.add("key", key);
        map.add("requestId", UUID.randomUUID().toString());
        map.add("timestamp", String.valueOf(System.currentTimeMillis()));
        HttpEntity<MultiValueMap<String, String>> request = new HttpEntity<>(map, headers);

        String value = restTemplate.postForObject("http://www.weiquding.com:8083/hello", request, String.class);
        LOGGER.info("receive value:[{}]", value);
        return value;
    }
}

  1. 准备一个服务端 RestfulApi 服务接口,并将服务端部署在远程主机上,此情况下不启动应用,另要确保远端防火墙相应的端口 8083 已经打开。
/**
 * java.net.SocketException : Connection reset 测试服务端
 */
@SpringBootApplication
@RestController
public class ApplicationCenter {

    private static final Logger LOGGER = LoggerFactory.getLogger(ApplicationCenter.class);

    public static void main(String[] args) {
        SpringApplication.run(ApplicationCenter.class, args);
    }

    @RequestMapping(path = "/hello", produces = MediaType.APPLICATION_JSON_VALUE)
    public String serverHello(HttpServletRequest request) {
        Enumeration<String> headerNames = request.getHeaderNames();
        while (headerNames.hasMoreElements()) {
            String headerName = headerNames.nextElement();
            LOGGER.info("headerName:[{}]==>{}", headerName, request.getHeader(headerName));
        }
        Enumeration<String> parameterNames = request.getParameterNames();
        while (parameterNames.hasMoreElements()){
            String parameterName = parameterNames.nextElement();
            LOGGER.info("parameterName:[{}]==>{}", parameterName, request.getParameter(parameterName));
        }
        return "server hello";
    }

}

  1. 在客户端打开 wireshark 软件,选择对应的数据交换的网络接口,在显示过滤器栏位输入规则"ip.addr == xxx.xxx.xxx.xxx and tcp.port==8083",以便只显示对 socket 上的流量。
  2. 启动客户端程序,调用远端服务。产生服务调用的方式为在本地浏览器输入 http://localhost:8082/hello?key=client
  3. 在 wireshark 的 Packet List 面板,查看双方交互的数据包。
    20191028160216.png
  4. 打开客户端控制台,查看报错的堆栈信息。
java.net.ConnectException: Connection refused: connect
	at java.base/java.net.DualStackPlainSocketImpl.connect0(Native Method) ~[na:na]
	at java.base/java.net.DualStackPlainSocketImpl.socketConnect(DualStackPlainSocketImpl.java:79) ~[na:na]
	at java.base/java.net.AbstractPlainSocketImpl.doConnect(AbstractPlainSocketImpl.java:400) ~[na:na]
	at java.base/java.net.AbstractPlainSocketImpl.connectToAddress(AbstractPlainSocketImpl.java:243) ~[na:na]
	at java.base/java.net.AbstractPlainSocketImpl.connect(AbstractPlainSocketImpl.java:225) ~[na:na]
	at java.base/java.net.PlainSocketImpl.connect(PlainSocketImpl.java:148) ~[na:na]
	at java.base/java.net.Socket.connect(Socket.java:591) ~[na:na]
	at java.base/java.net.Socket.connect(Socket.java:540) ~[na:na]
	at java.base/sun.net.NetworkClient.doConnect(NetworkClient.java:182) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.openServer(HttpClient.java:474) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.openServer(HttpClient.java:569) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.<init>(HttpClient.java:242) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.New(HttpClient.java:341) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.New(HttpClient.java:362) ~[na:na]
	at java.base/sun.net.www.protocol.http.HttpURLConnection.getNewHttpClient(HttpURLConnection.java:1242) ~[na:na]
	at java.base/sun.net.www.protocol.http.HttpURLConnection.plainConnect0(HttpURLConnection.java:1181) ~[na:na]
	at java.base/sun.net.www.protocol.http.HttpURLConnection.plainConnect(HttpURLConnection.java:1075) ~[na:na]
	at java.base/sun.net.www.protocol.http.HttpURLConnection.connect(HttpURLConnection.java:1009) ~[na:na]
	at org.springframework.http.client.SimpleBufferingClientHttpRequest.executeInternal(SimpleBufferingClientHttpRequest.java:76) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.http.client.AbstractBufferingClientHttpRequest.executeInternal(AbstractBufferingClientHttpRequest.java:48) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.http.client.AbstractClientHttpRequest.execute(AbstractClientHttpRequest.java:53) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.web.client.RestTemplate.doExecute(RestTemplate.java:742) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.web.client.RestTemplate.execute(RestTemplate.java:677) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.web.client.RestTemplate.postForObject(RestTemplate.java:421) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at com.weiquding.connection.reset.demo.client.ApplicationCenter.serverHello(ApplicationCenter.java:53) ~[classes/:na]
分析
  1. 由上节 wireshark 捕获的数据包可知,客户端与服务端只存在一次有效交互,即客户端发送 SYN 报文段,试图建立连接,但服务端由于没有端口监听,回了 RST 报文段。
  2. 由上节报错的堆栈信息可知,客户端建立连接的过程最终会调用到 Native 方程 connect0,以 windows 平台为例,方法为 java.net.DualStackPlainSocketImpl#connect0,对应源码 PlainSocketImpl#connect0 如下:
/*
 * Class:     java_net_PlainSocketImpl
 * Method:    connect0
 * Signature: (ILjava/net/InetAddress;I)I
 */
JNIEXPORT jint JNICALL Java_java_net_PlainSocketImpl_connect0
  (JNIEnv *env, jclass clazz, jint fd, jobject iaObj, jint port) {
    SOCKETADDRESS sa;
    int rv, sa_len = 0;
    jboolean v4MappedAddress = ipv6_available() ? JNI_TRUE : JNI_FALSE;

    if (NET_InetAddressToSockaddr(env, iaObj, port, &sa,
                                  &sa_len, v4MappedAddress) != 0) {
        return -1;
    }

    rv = connect(fd, &sa.sa, sa_len);
    if (rv == SOCKET_ERROR) {
        int err = WSAGetLastError();
        if (err == WSAEWOULDBLOCK) {
            return java_net_PlainSocketImpl_WOULDBLOCK;
        } else if (err == WSAEADDRNOTAVAIL) {
            JNU_ThrowByName(env, JNU_JAVANETPKG "ConnectException",
                "connect: Address is invalid on local machine,"
                " or port is not valid on remote machine");
        } else {
            /*
            * 当errno为WSAECONNREFUSED时,走此分支,抛出异常:java/net/ConnectException:Connection refused: connect
            *{ WSAECONNREFUSED,          "ConnectException",     "Connection refused" },
            */
            NET_ThrowNew(env, err, "connect");
        }
        // return value not important.
    }
    return rv;
}
  1. 此种情况下,异常在 TCP 建立过程中发生,抛出异常不是 java.net.SocketException:Connection reset。

检测半开连接

重现步骤
  1. 准备一个客户端 httpClient 请求示例。同上一情况。
  2. 准备一个服务端 RestfulApi 服务接口,并将服务端部署在远程主机上,此情况下先启动应用,等客户端执行 read 或 write 前再强制杀掉服务端应用,另要确保远端防火墙相应的端口 8083 已经打开。
  3. 在客户端打开 wireshark 软件,选择对应的数据交换的网络接口(此时选择本地回环网络接口 Loopback,因为因特网半开连接不好模拟),在显示过滤器栏位输入规则"ip.addr == xxx.xxx.xxx.xxx and tcp.port==8083",以便只显示对 socket 上的流量。
客户端 write 时遇到半开连接的情况
  1. 首先看 write 时遇到半开连接的情况,在 java.net.SocketOutputStream#socketWrite 方法的 socketWrite0 方法调用处打一个断点。
/**
     * Writes to the socket with appropriate locking of the
     * FileDescriptor.
     * @param b the data to be written
     * @param off the start offset in the data
     * @param len the number of bytes that are written
     * @exception IOException If an I/O error has occurred.
     */
    private void socketWrite(byte b[], int off, int len) throws IOException {


        if (len <= 0 || off < 0 || len > b.length - off) {
            if (len == 0) {
                return;
            }
            throw new ArrayIndexOutOfBoundsException("len == " + len
                    + " off == " + off + " buffer length == " + b.length);
        }

        FileDescriptor fd = impl.acquireFD();
        try {
            // 以下行为断点位置
            socketWrite0(fd, b, off, len);
        } catch (SocketException se) {
            /* 此处会抛出异常:java.net.SocketException: Connection reset by peer: socket write error
            * 但由于整个out.flush过程会调用三次,且异常发生后存在半开连接的拆除和重连接。
            * 最终的异常信息为:java.net.ConnectException: Connection refused: connect
            *
            */
            if (se instanceof sun.net.ConnectionResetException) {
                impl.setConnectionResetPending();
                se = new SocketException("Connection reset");
            }
            if (impl.isClosedOrPending()) {
                throw new SocketException("Socket closed");
            } else {
                // 异常抛出之处
                throw se;
            }
        } finally {
            impl.releaseFD();
        }
    }
  1. 启动客户端程序,调用远端服务。产生服务调用的方式为在本地浏览器输入 http://localhost:8082/hello?key=client

  2. 当客户端程序运行到 socketWrite0 时,强关服务端程序(idea 中直接关停应用即可),然后再放开断点让客户端程序执行完成。此时客户端应先已经产生异常:java.net.SocketException: Connection reset by peer: socket write error,但不抛出,这部分原因在分析章节指出。

  3. . 在 wireshark 的 Packet List 面板,查看双方交互的数据包。20191028182803.png

  4. 打开客户端控制台,查看报错的堆栈信息。

java.net.ConnectException: Connection refused: connect
	at java.base/java.net.DualStackPlainSocketImpl.connect0(Native Method) ~[na:na]
	at java.base/java.net.DualStackPlainSocketImpl.socketConnect(DualStackPlainSocketImpl.java:79) ~[na:na]
	at java.base/java.net.AbstractPlainSocketImpl.doConnect(AbstractPlainSocketImpl.java:400) ~[na:na]
	at java.base/java.net.AbstractPlainSocketImpl.connectToAddress(AbstractPlainSocketImpl.java:243) ~[na:na]
	at java.base/java.net.AbstractPlainSocketImpl.connect(AbstractPlainSocketImpl.java:225) ~[na:na]
	at java.base/java.net.PlainSocketImpl.connect(PlainSocketImpl.java:148) ~[na:na]
	at java.base/java.net.Socket.connect(Socket.java:591) ~[na:na]
	at java.base/java.net.Socket.connect(Socket.java:540) ~[na:na]
	at java.base/sun.net.NetworkClient.doConnect(NetworkClient.java:182) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.openServer(HttpClient.java:474) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.openServer(HttpClient.java:569) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.<init>(HttpClient.java:242) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.New(HttpClient.java:341) ~[na:na]
	at java.base/sun.net.www.http.HttpClient.New(HttpClient.java:376) ~[na:na]
	at java.base/sun.net.www.protocol.http.HttpURLConnection.setNewClient(HttpURLConnection.java:793) ~[na:na]
	at java.base/sun.net.www.protocol.http.HttpURLConnection.setNewClient(HttpURLConnection.java:781) ~[na:na]
	at java.base/sun.net.www.protocol.http.HttpURLConnection.writeRequests(HttpURLConnection.java:723) ~[na:na]
	at java.base/sun.net.www.protocol.http.HttpURLConnection.getOutputStream0(HttpURLConnection.java:1366) ~[na:na]
	at java.base/sun.net.www.protocol.http.HttpURLConnection.getOutputStream(HttpURLConnection.java:1331) ~[na:na]
	at org.springframework.http.client.SimpleBufferingClientHttpRequest.executeInternal(SimpleBufferingClientHttpRequest.java:78) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.http.client.AbstractBufferingClientHttpRequest.executeInternal(AbstractBufferingClientHttpRequest.java:48) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.http.client.AbstractClientHttpRequest.execute(AbstractClientHttpRequest.java:53) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.web.client.RestTemplate.doExecute(RestTemplate.java:742) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.web.client.RestTemplate.execute(RestTemplate.java:677) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at org.springframework.web.client.RestTemplate.postForObject(RestTemplate.java:421) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
	at com.weiquding.connection.reset.demo.client.ApplicationCenter.serverHello(ApplicationCenter.java:53) ~[classes/:na]
客户端 write 时遇到半开连接情况的分析
  1. 客户端在 java.net.SocketOutputStream#socketWrite 方法中得到异常信息 java.net.SocketException: Connection reset by peer: socket write error 后,为什么没有抛出?
    • 由于存在整个 PrintStream-->BufferedOutputStream-->SocketOutputStream 包装,out.flush 过程会调用三次 socketWirte,且异常发生后异常信息会捕获,后续进行半开连接的拆除和重连接。见异常捕获 PrintStream.flush()方法及三次 socketWrite 调用及连接重建的的栈信息。
      /**
       * Flushes the stream.  This is done by writing any buffered output bytes to
       * the underlying output stream and then flushing that stream.
       *
       * @see        java.io.OutputStream#flush()
       */
      public void flush() {
          synchronized (this) {
              try {
                  ensureOpen();
                  //此处调用BufferedOutputStream.flush();间接调用SocketOutputStream.socketWrite。
                  out.flush();
              }
              catch (IOException x) {
                  // 异常捕获,设置异常标志。
                  trouble = true;
              }
          }
      }
      
    • 第一次调用 socketWrite 方法的栈信息。
      socketWrite:111, SocketOutputStream (java.net)
      write:155, SocketOutputStream (java.net)
      flushBuffer:81, BufferedOutputStream (java.io)
      flush:142, BufferedOutputStream (java.io)
      flush:417, PrintStream (java.io)
      print:301, MessageHeader (sun.net.www)
      writeRequests:655, HttpClient (sun.net.www.http)
      writeRequests:666, HttpClient (sun.net.www.http)
      writeRequests:711, HttpURLConnection (sun.net.www.protocol.http)
      getOutputStream0:1366, HttpURLConnection (sun.net.www.protocol.http)
      getOutputStream:1331, HttpURLConnection (sun.net.www.protocol.http)
      executeInternal:78, SimpleBufferingClientHttpRequest (org.springframework.http.client)
      executeInternal:48, AbstractBufferingClientHttpRequest (org.springframework.http.client)
      execute:53, AbstractClientHttpRequest (org.springframework.http.client)
      doExecute:742, RestTemplate (org.springframework.web.client)
      execute:677, RestTemplate (org.springframework.web.client)
      postForObject:421, RestTemplate (org.springframework.web.client)
      serverHello:53, ApplicationCenter (com.weiquding.connection.reset.demo.client)
      
    • HttpClient#writeRequests(sun.net.www.MessageHeader, sun.net.www.http.PosterOutputStream)方法
      public void writeRequests(MessageHeader head,
                            PosterOutputStream pos) throws IOException {
          requests = head;
          // 第一次调用SocketOutputStream.socketWrite的入口
          requests.print(serverOutput);
          poster = pos;
          if (poster != null)
              poster.writeTo(serverOutput);
          // PrintStream.flush, 第二次调用SocketOutputStream.socketWrite的入口
          serverOutput.flush();
      }
      
    • 第二调用 SocketOutputStream.socketWrite 的栈信息
      socketWrite:111, SocketOutputStream (java.net)
      write:155, SocketOutputStream (java.net)
      flushBuffer:81, BufferedOutputStream (java.io)
      flush:142, BufferedOutputStream (java.io)
      flush:417, PrintStream (java.io)
      writeRequests:659, HttpClient (sun.net.www.http)
      writeRequests:666, HttpClient (sun.net.www.http)
      writeRequests:711, HttpURLConnection (sun.net.www.protocol.http)
      getOutputStream0:1366, HttpURLConnection (sun.net.www.protocol.http)
      getOutputStream:1331, HttpURLConnection (sun.net.www.protocol.http)
      executeInternal:78, SimpleBufferingClientHttpRequest (org.springframework.http.client)
      executeInternal:48, AbstractBufferingClientHttpRequest (org.springframework.http.client)
      execute:53, AbstractClientHttpRequest (org.springframework.http.client)
      doExecute:742, RestTemplate (org.springframework.web.client)
      execute:677, RestTemplate (org.springframework.web.client)
      postForObject:421, RestTemplate (org.springframework.web.client)
      serverHello:53, ApplicationCenter (com.weiquding.connection.reset.demo.client)
      
    • java.io.PrintStream#checkError 方法
      /**
       * Flushes the stream and checks its error state. The internal error state
       * is set to {@code true} when the underlying output stream throws an
       * {@code IOException} other than {@code InterruptedIOException},
       * and when the {@code setError} method is invoked.  If an operation
       * on the underlying output stream throws an
       * {@code InterruptedIOException}, then the {@code PrintStream}
       * converts the exception back into an interrupt by doing:
       * <pre>{@code
       *     Thread.currentThread().interrupt();
       * }</pre>
       * or the equivalent.
       *
       * @return {@code true} if and only if this stream has encountered an
       *         {@code IOException} other than
       *         {@code InterruptedIOException}, or the
       *         {@code setError} method has been invoked
       */
      public boolean checkError() {
          if (out != null)
              // PrintStream.flush, 第三次调用SocketOutputStream.socketWrite的入口
              flush();
          if (out instanceof java.io.PrintStream) {
              PrintStream ps = (PrintStream) out;
              return ps.checkError();
          }
          // 返回是否异常,true将进入连接重建逻辑。
          return trouble;
      }
      
    • 第三次调用 SocketOutputStream.socketWrite 的栈信息
      socketWrite:111, SocketOutputStream (java.net)
      write:155, SocketOutputStream (java.net)
      flushBuffer:81, BufferedOutputStream (java.io)
      flush:142, BufferedOutputStream (java.io)
      flush:417, PrintStream (java.io)
      checkError:471, PrintStream (java.io)
      writeRequests:712, HttpURLConnection (sun.net.www.protocol.http)
      getOutputStream0:1366, HttpURLConnection (sun.net.www.protocol.http)
      getOutputStream:1331, HttpURLConnection (sun.net.www.protocol.http)
      executeInternal:78, SimpleBufferingClientHttpRequest (org.springframework.http.client)
      executeInternal:48, AbstractBufferingClientHttpRequest (org.springframework.http.client)
      execute:53, AbstractClientHttpRequest (org.springframework.http.client)
      doExecute:742, RestTemplate (org.springframework.web.client)
      execute:677, RestTemplate (org.springframework.web.client)
      postForObject:421, RestTemplate (org.springframework.web.client)
      serverHello:53, ApplicationCenter (com.weiquding.connection.reset.demo.client)
      
    • sun.net.www.protocol.http.HttpURLConnection#writeRequests 重建 TCP 链接逻辑分析
      /* adds the standard key/val pairs to reqests if necessary & write to
       * given PrintStream
       */
      private void writeRequests() throws IOException {
          /*
          * 省略设计http请求首部的逻辑
          */
      
          // 前两次的SocketOutputStream.socketWrite调用入口
          http.writeRequests(requests, poster, streaming());
          // 第三次SocketOutputStream.socketWrite调用,并检查是否有异常
          if (ps.checkError()) {
              String proxyHost = http.getProxyHostUsed();
              int proxyPort = http.getProxyPortUsed();
              // 断开半开连接
              disconnectInternal();
              if (failedOnce) {
                  throw new IOException("Error writing to server");
              } else { // try once more
                  failedOnce=true;
                  if (proxyHost != null) {
                      setProxiedClient(url, proxyHost, proxyPort);
                  } else {
                      // 进行一次连接重建,此时会抛出java.net.ConnectException: Connection refused: connect
                      setNewClient (url);
                  }
                  ps = (PrintStream) http.getOutputStream();
                  connected=true;
                  responses = new MessageHeader();
                  setRequests=false;
                  // 递归调用,以便重新进入socketWrite调用
                  writeRequests();
              }
          }
      }
      
  2. 从上述源码和栈调用可以看出,HttpURLConnection 在处理 write 时进行很好的容错处理,包括以下两点:
    • 如果缓冲区仍存在数据,多次尝试写入 socket.
    • 多次写失败后,尝试重建连接,如果新的连接建立成功,再次开始写入流程。
客户端 read 时遇到半开连接情况
  1. 前三步与 wirte 时一致,不再叙述。对于 read 时遇到半开连接的情况,在 java.net.SocketInputStream#read(byte[], int, int, int)方法的 socketRead 方法调用处打一个断点。

    /**
         * Reads into a byte array <i>b</i> at offset <i>off</i>,
         * <i>length</i> bytes of data.
         * @param b the buffer into which the data is read
         * @param off the start offset of the data
         * @param length the maximum number of bytes read
         * @return the actual number of bytes read, -1 is
         *          returned when the end of the stream is reached.
         * @exception IOException If an I/O error has occurred.
         */
        public int read(byte b[], int off, int length) throws IOException {
            return read(b, off, length, impl.getTimeout());
        }
    
        int read(byte b[], int off, int length, int timeout) throws IOException {
            int n;
    
            // EOF already encountered
            if (eof) {
                return -1;
            }
    
            // connection reset
            if (impl.isConnectionReset()) {
                throw new SocketException("Connection reset");
            }
    
            // bounds check
            if (length <= 0 || off < 0 || length > b.length - off) {
                if (length == 0) {
                    return 0;
                }
                throw new ArrayIndexOutOfBoundsException("length == " + length
                        + " off == " + off + " buffer length == " + b.length);
            }
    
            boolean gotReset = false;
    
            // acquire file descriptor and do the read
            FileDescriptor fd = impl.acquireFD();
            try {
                // 以下行为打断点处,此方法将调用Native方法socketRead0
                n = socketRead(fd, b, off, length, timeout);
                if (n > 0) {
                    return n;
                }
            } catch (ConnectionResetException rstExc) {
                // 捕获sun.net.ConnectionResetException连接重置异常,改为设置标志以便再次尝试读取缓存区数据
                gotReset = true;
            } finally {
                impl.releaseFD();
            }
    
            /*
             * We receive a "connection reset" but there may be bytes still
             * buffered on the socket
             * 再次尝试读取缓存区数据
             */
            if (gotReset) {
                impl.setConnectionResetPending();
                impl.acquireFD();
                try {
                    n = socketRead(fd, b, off, length, timeout);
                    if (n > 0) {
                        return n;
                    }
                } catch (ConnectionResetException rstExc) {
                } finally {
                    impl.releaseFD();
                }
            }
    
            /*
             * If we get here we are at EOF, the socket has been closed,
             * or the connection has been reset.
             */
            if (impl.isClosedOrPending()) {
                throw new SocketException("Socket closed");
            }
            if (impl.isConnectionResetPending()) {
                impl.setConnectionReset();
            }
            if (impl.isConnectionReset()) {
                // 如果两次都因为连接重置没有读取成功,转换异常,抛出java.net.SocketException: Connection reset异常。
                throw new SocketException("Connection reset");
            }
            eof = true;
            return -1;
        }
    
    
  2. 启动客户端程序,调用远端服务。产生服务调用的方式为在本地浏览器输入 http://localhost:8082/hello?key=client

  3. 当客户端程序运行到 socketRead 时,强关服务端程序(idea 中直接关停应用即可),然后再放开断点让客户端程序执行完成。此时客户端产出异常:sun.net.ConnectionResetException。

  4. 在 wireshark 的 Packet List 面板,查看双方交互的数据包。20191028205230.png

  5. 打开客户端控制台,查看报错的堆栈信息。

    java.net.SocketException: Connection reset
    	at java.base/java.net.SocketInputStream.read(SocketInputStream.java:210) ~[na:na]
    	at java.base/java.net.SocketInputStream.read(SocketInputStream.java:141) ~[na:na]
    	at java.base/java.io.BufferedInputStream.fill(BufferedInputStream.java:252) ~[na:na]
    	at java.base/java.io.BufferedInputStream.read1(BufferedInputStream.java:292) ~[na:na]
    	at java.base/java.io.BufferedInputStream.read(BufferedInputStream.java:351) ~[na:na]
    	at java.base/sun.net.www.http.HttpClient.parseHTTPHeader(HttpClient.java:746) ~[na:na]
    	at java.base/sun.net.www.http.HttpClient.parseHTTP(HttpClient.java:689) ~[na:na]
    	at java.base/sun.net.www.protocol.http.HttpURLConnection.getInputStream0(HttpURLConnection.java:1604) ~[na:na]
    	at java.base/sun.net.www.protocol.http.HttpURLConnection.getInputStream(HttpURLConnection.java:1509) ~[na:na]
    	at java.base/java.net.HttpURLConnection.getResponseCode(HttpURLConnection.java:527) ~[na:na]
    	at org.springframework.http.client.SimpleClientHttpResponse.getRawStatusCode(SimpleClientHttpResponse.java:55) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
    	at org.springframework.web.client.DefaultResponseErrorHandler.hasError(DefaultResponseErrorHandler.java:55) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
    	at org.springframework.web.client.RestTemplate.handleResponse(RestTemplate.java:773) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
    	at org.springframework.web.client.RestTemplate.doExecute(RestTemplate.java:743) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
    	at org.springframework.web.client.RestTemplate.execute(RestTemplate.java:677) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
    	at org.springframework.web.client.RestTemplate.postForObject(RestTemplate.java:421) ~[spring-web-5.2.0.RELEASE.jar:5.2.0.RELEASE]
    	at com.weiquding.connection.reset.demo.client.ApplicationCenter.serverHello(ApplicationCenter.java:53) ~[classes/:na]
    
客户端 read 时遇到半开连接情况的分析
  1. read 产生 Connection reset 需要在第一次进行 socketRead 时服务端已经回了 RST 报文段。
  2. 从 SocketInputStream.read 方法的源码中,可以在异常发生后还会尝试进行一次重试,考虑了容错性。

异常终止一个连接

重现步骤
  1. 由于没有找到 tomcat 直接没有设置 Socket 参数 SO_LINGER 的地方,写一对 Socket 客户端和服务端来模拟此情形。

  2. 写一个 Socket 客户端。示例代码见参考资料的 connetion-reset-demo。

        @RequestMapping(path = "/socket", produces = MediaType.APPLICATION_JSON_VALUE)
        public String connectSocket(@RequestParam("key")String key) {
            Socket socket = null;
            InputStream inputStream = null;
            OutputStream outputStream = null;
            try {
                socket = new Socket();
                socket.connect(new InetSocketAddress("127.0.0.1", 8084));
                outputStream = socket.getOutputStream();
                outputStream.write(key.getBytes(StandardCharsets.UTF_8));
                inputStream = socket.getInputStream();
                byte[] rcvBytes = new byte[1024];
                inputStream.read(rcvBytes);
                return new String(rcvBytes, StandardCharsets.UTF_8);
            }catch (IOException e){
                LOGGER.error("socket通信异常", e);
            }finally {
                if(outputStream != null){
                    try {
                        outputStream.close();
                    }catch (IOException e){
                        LOGGER.error("关闭outputStream异常", e);
                    }
                }
                if(inputStream != null){
                    try {
                        inputStream.close();
                    }catch (IOException e){
                        LOGGER.error("关闭inputStream异常", e);
                    }
                }
                if(socket != null){
                    try {
                        socket.close();
                    } catch (IOException e) {
                        LOGGER.error("关闭socket异常", e);
                    }
                }
            }
            return "";
        }
    }
    
  3. 写一个 ServerSocket 服务端,设置 SO_LINGER 参数

    public class EchoServer {
    
        private static final Logger LOGGER = LoggerFactory.getLogger(EchoServer.class);
    
    
        public static void main(String[] args) {
            try(ServerSocket serverSocket = new ServerSocket(8084)) {
                while(true){
                    Socket accept = serverSocket.accept();
                    new Thread(new ThreadHandler(accept)).start();
                }
            } catch (IOException e) {
                LOGGER.error("serverSocket异常", e);
            }
        }
    
        static class ThreadHandler implements Runnable{
    
            private Socket socket;
    
            public ThreadHandler(Socket socket){
                this.socket = socket;
            }
    
            @Override
            public void run() {
                InputStream inputStream = null;
                OutputStream outputStream = null;
                try {
                    // 设置SO_LINGER
                    socket.setSoLinger(true, 0);
                    inputStream = socket.getInputStream();
                    byte[] rcvBytes = new byte[1024];
                    inputStream.read(rcvBytes);
                    LOGGER.info("receive data:[{}]", new String(rcvBytes, StandardCharsets.UTF_8));
                    outputStream = socket.getOutputStream();
                    outputStream.write("server socket".getBytes(StandardCharsets.UTF_8));
                    socket.close();
                } catch (IOException e) {
                   LOGGER.error("处理socket异常", e);
                }finally {
                    if(inputStream != null){
                        try {
                            inputStream.close();
                        }catch (IOException e){
                            LOGGER.error("关闭inputStream异常", e);
                        }
                    }
                    if(outputStream != null){
                        try {
                            outputStream.close();
                        }catch (IOException e){
                            LOGGER.error("关闭outputStream异常", e);
                        }
                    }
                    try {
                        socket.close();
                    } catch (IOException e) {
                        LOGGER.error("关闭socket异常", e);
                    }
                }
    
            }
        }
    }
    
  4. 启动服务端进行监听

  5. 在客户端打开 wireshark 软件,选择对应的数据交换的网络接口(此时选择本地回环网络接口 Loopback,因为因特网半开连接不好模拟),在显示过滤器栏位输入规则"ip.addr == xxx.xxx.xxx.xxx and tcp.port==8084",以便只显示对应 socket 上的流量。

  6. 启动客户端,通过调用 http://localhost:8082/socket?key=client,产生服务端 socket 通信。

  7. 在 wireshark 的 Packet List 面板,查看双方交互的数据包。20191028222805.png

  8. 打开客户端控制台,查看报错的堆栈信息。

java.net.SocketException: Connection reset
	at java.base/java.net.SocketInputStream.read(SocketInputStream.java:210) ~[na:na]
	at java.base/java.net.SocketInputStream.read(SocketInputStream.java:141) ~[na:na]
	at java.base/java.net.SocketInputStream.read(SocketInputStream.java:127) ~[na:na]
	at com.weiquding.connection.reset.demo.client.ApplicationCenter.connectSocket(ApplicationCenter.java:82) ~[classes/:na]
分析
  1. 服务端设置了 SO_LINGER,关闭链接不会发 FIN 报文段,而是发 RST 报文段。
  2. 客户端收到此 RST 报文段,再进行 SocketInputStream.read,将抛出 java.net.SocketException: Connection reset。

解决故障的方案

  1. 对于“到不存在的端口的连接请求”的情况。
    • 如果是程序端口错误,无法解决。
    • 其它情况可以在设置服务端多实例,客户端使用降级重试策略,提高可用性。
  2. 对于“检测半开连接”的情况。
    • 可以使用 Http 短连接,减少长连接使用过程服务端发生异常导致半开连接的机会,但这会降低性能。
    • 客户端提供 Http 连接池,单个连接使用长连接并配置较短的超时时间,对连接复用,能减少连接创建的消耗;同时连接池提供连接有效性的检测,有无效连接进行拆除。
    • 如果服务端只有单一实例,其反复出现应用崩溃等引起半开连接,则要检测应用程序是否有问题。另外要考虑服务端多实例部署,并在客户端准备降级策略。
  3. 对于“异常终止一个连接”的情况。
    • 服务端不要设置 Socket 参数 SO_LINGER,执行正常的连接终止流程。同时连接的终止应当考虑由客户端发起。

参考资料

  1. Setting the SO_LINGER Option
  2. TCP option SO_LINGER (zero) - when it's required
  3. What's causing my java.net.SocketException: Connection reset?
  4. java.net.SocketException: Connection reset
  5. Understanding some common SocketExceptions in JAVA
  6. Tomcat 配置 api
  7. TCP/IP 详解.卷一 P187 异常中止一个连接
  8. HTTP 权威指南 P100 Keep-Alive 和哑代理
  9. 计算机网络(自顶向下方法)P170 TCP 连接管理
  10. Wireshark 数据包分析实战 P117 TCP 重置
  11. TIME_WAIT and its design implications for protocols and scalable client server systems
  12. connection-reset-demo 示例程序地址
  13. Detection of Half-Open (Dropped) Connections
  • Java

    Java 是一种可以撰写跨平台应用软件的面向对象的程序设计语言,是由 Sun Microsystems 公司于 1995 年 5 月推出的。Java 技术具有卓越的通用性、高效性、平台移植性和安全性。

    3167 引用 • 8207 回帖
  • 网络
    128 引用 • 177 回帖 • 3 关注
  • Wireshark
    2 引用 • 1 回帖

相关帖子

欢迎来到这里!

我们正在构建一个小众社区,大家在这里相互信任,以平等 • 自由 • 奔放的价值观进行分享交流。最终,希望大家能够找到与自己志同道合的伙伴,共同成长。

注册 关于
请输入回帖内容 ...
  • adlered

    标题错误:

    java.net.SoecketException => java.net.SocketException