【发布时间】:2015-03-26 09:33:48
【问题描述】:
我有一个项目,在该项目中,我在许多任务中同时下载了许多页面,这些页面通过ThreadPool(大小 = 200)进行处理。所有这些任务都使用相同的方法getPage 下载页面(使用 Apache Commons HttpClient 和 Apache Commons IO):
public static String getPage(String url)
throws IOException {
HttpUriRequest request = new HttpGet(url);
HttpResponse response = HTTP_CLIENT_BUILDER.build().execute(request);
try (InputStream content = response.getEntity().getContent()) {
return IOUtils.toString(content, "UTF-8");
}
}
而HTTP_CLIENT_BUILDER 是这样初始化的静态字段:
private static final HttpClientBuilder HTTP_CLIENT_BUILDER = HttpClients.custom()
.setDefaultRequestConfig(RequestConfig.custom()
.setSocketTimeout(SOCKET_TIMEOUT_MS) // 60_000
.setConnectTimeout(CONNECTION_TIMEOUT_MS) // 5_000
.build());
问题陈述:在某个时刻(大部分任务完成时)所有剩余线程都卡在本机方法SocketInputStream.socketRead0,所以jdb 说,他们'重新运行(嗯,是的,我希望本机方法运行时的行为:-)):
> threads
Group system:
(java.lang.ref.Reference$ReferenceHandler)0xac4 Reference Handler cond. waiting
(java.lang.ref.Finalizer$FinalizerThread)0xac5 Finalizer cond. waiting
(java.lang.Thread)0xac6 Signal Dispatcher running
(java.lang.Thread)0xac7 Java2D Disposer cond. waiting
Group main:
(java.lang.Thread)0xac9 pool-1-thread-5 running
(java.lang.Thread)0xaca pool-1-thread-12 running
(... 12 more threads from ThreadPool ...)
(java.lang.Thread)0xad7 DestroyJavaVM running
> where 0xac9
[1] java.net.SocketInputStream.socketRead0 (native method)
[2] java.net.SocketInputStream.read (SocketInputStream.java:150)
[3] java.net.SocketInputStream.read (SocketInputStream.java:121)
[4] sun.security.ssl.InputRecord.readFully (InputRecord.java:465)
[5] sun.security.ssl.InputRecord.read (InputRecord.java:503)
[6] sun.security.ssl.SSLSocketImpl.readRecord (SSLSocketImpl.java:961)
[7] sun.security.ssl.SSLSocketImpl.performInitialHandshake (SSLSocketImpl.java:1,363)
[8] sun.security.ssl.SSLSocketImpl.startHandshake (SSLSocketImpl.java:1,391)
[9] sun.security.ssl.SSLSocketImpl.startHandshake (SSLSocketImpl.java:1,375)
[10] org.apache.http.conn.ssl.SSLConnectionSocketFactory.createLayeredSocket (SSLConnectionSocketFactory.java:275)
[11] org.apache.http.conn.ssl.SSLConnectionSocketFactory.connectSocket (SSLConnectionSocketFactory.java:254)
[12] org.apache.http.impl.conn.HttpClientConnectionOperator.connect (HttpClientConnectionOperator.java:117)
[13] org.apache.http.impl.conn.PoolingHttpClientConnectionManager.connect (PoolingHttpClientConnectionManager.java:314)
[14] org.apache.http.impl.execchain.MainClientExec.establishRoute (MainClientExec.java:363)
[15] org.apache.http.impl.execchain.MainClientExec.execute (MainClientExec.java:219)
[16] org.apache.http.impl.execchain.ProtocolExec.execute (ProtocolExec.java:195)
[17] org.apache.http.impl.execchain.RetryExec.execute (RetryExec.java:86)
[18] org.apache.http.impl.execchain.RedirectExec.execute (RedirectExec.java:108)
[19] org.apache.http.impl.client.InternalHttpClient.doExecute (InternalHttpClient.java:186)
[20] org.apache.http.impl.client.CloseableHttpClient.execute (CloseableHttpClient.java:82)
[21] org.apache.http.impl.client.CloseableHttpClient.execute (CloseableHttpClient.java:106)
[22] <package>.Utils.getPage (Utils.java:122)
[23...] <internal details>
> # the same picture for all of them
我不明白为什么会发生这种情况,但我找到了Java bug,这可能与问题有关。所以也许我不是在寻找真正的解决方案,而是在寻找一些解决方法。
由于该错误是针对 Linux 提交的,我应该说,我也在使用运行 Ubuntu 14.04 x86_64 的虚拟机
UPD:好的,我现在尝试使用setConnectionRequestTimeout 添加新的超时(只是为了确保它不起作用)添加finally 阻止getPage:
...
try (InputStream content = response.getEntity().getContent()) {
return IOUtils.toString(content, "UTF-8");
} finally {
httpClient.getConnectionManager().closeIdleConnections(0, TimeUnit.NANOSECONDS);
}
让我们看看,这是否有帮助。
UPD2:这似乎有点帮助,但我仍然有这个永久运行的任务大约每天卡住一次。
【问题讨论】:
-
执行
Socket.read的线程将显示为Runnable,请参阅此SO 帖子:stackoverflow.com/questions/12544212。最有可能的是,远程端保持其套接字末端打开,这就是您的任务无法完成的原因。例如,您向 Executor 提交的任务多于下载远程资源所需的任务,而其余空闲任务则处于等待状态。 -
@VictorSorokin 这不应该发生,因为我设置了超时(请参阅
HTTP_CLIENT_BUILDER的初始化程序) -
是的,忽略了,抱歉。然后我会检查与 tcpdump 或类似的连接,以了解是什么让 TCP 连接保持活动状态。也许,服务器端日志也可以很方便。
-
然后,一旦您知道(不知何故)所有内容都已下载,请关闭您身边打开的套接字。这将释放池线程并取消卡住的任务(有例外)。这意味着您需要将打开的套接字放在
getPage内的某个共享集合中,这些集合应该可用于取消任务。 -
@VictorSorokin 套接字本身是 HttpClient 引擎盖下的实现细节,所以我不确定,我怎样才能得到它们。此外,在我看来,创建这样一个集合有点过头了。
标签: java sockets apache-commons-httpclient