发现很多人搞爬虫会把python作为首选技术,理由是简单;但是本人最熟悉的还是java,所以对java内存浏览器技术htmlunit做了一次研究,发现原生的htmlunit的性能及对多线程的支持不是那么友好,特别是使用代理ip后,oom是很正常的,监控程序并查看源码总结问题原因:
- 1、js执行器执行js是使用多线程执行,在关闭js执行线程的时候,使用com.gargoylesoftware.htmlunit.javascript.background.DefaultJavaScriptExecutor这个类的时候,有段代码。
引用
private void killThread() {
if (eventLoopThread_ == null) {
return;
}
try {
eventLoopThread_.interrupt();
eventLoopThread_.join(10_000);
}
catch (final InterruptedException e) {
LOG.warn("InterruptedException while waiting for the eventLoop thread to join ", e);
// ignore, this doesn't matter, we want to stop it
}
if (eventLoopThread_.isAlive()) {
if (LOG.isWarnEnabled()) {
LOG.warn("Event loop thread "
+ eventLoopThread_.getName()
+ " still alive at "
+ System.currentTimeMillis());
LOG.warn("Event loop thread will be stopped");
}
// Stop the thread
eventLoopThread_.stop();
}
}
上面代码的问题:
引用
eventLoopThread_.interrupt();
eventLoopThread_.join(10_000);
不要觉得interrupt真的就会关闭线程,比如正在执行io操作或者同样该线程在处于sleep状态,interrupt就不会终止线程,所以这里住线程要等待eventLoopThread执行10秒才会继续往下跑。
2、DefaultJavaScriptExecutor在使用外部线程池开启webclient抓取网页的时候,经常会出现线程不关闭的情况,问题代码如下:
引用
public void run() {
final boolean trace = LOG.isTraceEnabled();
// this has to be a multiple of 10ms
// otherwise the VM has to fight with the OS to get such small periods
final long sleepInterval = 10;
while (!shutdown_.get() && !Thread.currentThread().isInterrupted() && webClient_.get() != null) {
final JavaScriptJobManager jobManager = getJobManagerWithEarliestJob();
if (jobManager != null) {
final JavaScriptJob earliestJob = jobManager.getEarliestJob();
if (earliestJob != null) {
final long waitTime = earliestJob.getTargetExecutionTime() - System.currentTimeMillis();
// do we have to execute the earliest job
if (waitTime < 1) {
// execute the earliest job
if (trace) {
LOG.trace("started executing job at " + System.currentTimeMillis());
}
jobManager.runSingleJob(earliestJob);
if (trace) {
LOG.trace("stopped executing job at " + System.currentTimeMillis());
}
// job is done, have a look for another one
continue;
}
}
}
// check for cancel
if (shutdown_.get() || Thread.currentThread().isInterrupted() || webClient_.get() == null) {
break;
}
// nothing to do, let's sleep a bit
try {
Thread.sleep(sleepInterval);
}
catch (final InterruptedException e) {
Thread.currentThread().interrupt();
}
}
}
此处问题代码:
引用
while (!shutdown_.get() && !Thread.currentThread().isInterrupted() && webClient_.get() != null)
外部线程内部关闭webclient.close()的时候,当外部线程要主动关闭本线程的时候,就像outStream没把out.close()写在finally里面,永远不会关闭js执行器线程。
- 3、其实htmlunit性能差还有一个最重要的问题所在,就是每次抓取同一个页面,都要去下载相同的资源,htmlunit下载页面的代码是在类com.gargoylesoftware.htmlunit.HttpWebConnection里面(js,css,jpg)
主要的方法代码如下:
引用
/**
* Reads the content of the stream and saves it in memory or on the file system.
* @param is the stream to read
* @param maxInMemory the maximumBytes to store in memory, after which save to a local file
* @return a wrapper around the downloaded content
* @throws IOException in case of read issues
*/
public static DownloadedContent downloadContent(final InputStream is, final int maxInMemory) throws IOException {
if (is == null) {
return new DownloadedContent.InMemory(null);
}
try (ByteArrayOutputStream bos = new ByteArrayOutputStream()) {
final byte[] buffer = new byte[1024];
int nbRead;
try {
while ((nbRead = is.read(buffer)) != -1) {
bos.write(buffer, 0, nbRead);
if (bos.size() > maxInMemory) {
// we have exceeded the max for memory, let's write everything to a temporary file
final File file = File.createTempFile("htmlunit", ".tmp");
file.deleteOnExit();
try (OutputStream fos = Files.newOutputStream(file.toPath())) {
bos.writeTo(fos); // what we have already read
IOUtils.copyLarge(is, fos); // what remains from the server response
}
return new DownloadedContent.OnFile(file, true);
}
}
}
catch (final ConnectionClosedException e) {
LOG.warn("Connection was closed while reading from stream.", e);
return new DownloadedContent.InMemory(bos.toByteArray());
}
catch (final EOFException e) {
// this might happen with broken gzip content
LOG.warn("EOFException while reading from stream.", e);
return new DownloadedContent.InMemory(bos.toByteArray());
}
return new DownloadedContent.InMemory(bos.toByteArray());
}
}
修改代码,只要把重复下载的代码缓存起来,就可以大大增加抓取性能,同时还可以动态修改网页js。
4、htmlunit设置requestTimeout时是无法单独设置conectiontimeout和socketTimeout,比方说设置requestTimeout=10000,那么 htmlclient的conectiontimeout=10000和socketTimeout=10000,这是有问题的,conectiontimeout一般情况应该设置低于100毫秒为宜,设置代码在 com.gargoylesoftware.htmlunit.HttpWebConnection
方法:
引用
private static RequestConfig.Builder createRequestConfigBuilder(final int timeout, final InetAddress localAddress) {
final RequestConfig.Builder requestBuilder = RequestConfig.custom()
.setCookieSpec(HACKED_COOKIE_POLICY)
.setRedirectsEnabled(false)
.setLocalAddress(localAddress)
// timeout
.setConnectTimeout(timeout)
.setConnectionRequestTimeout(timeout)
.setSocketTimeout(timeout);
return requestBuilder;
}
综上,把上述几点都完善,htmlunit不比只能多进程的python爬虫性能差,而且能够做黑帽。
已有 0 人发表留言,猛击->> 这里<<-参与讨论
ITeye推荐