I am scraping the content of web pages heavily in a multi-thread environment. I need a reliable downloader component that is tolerable to temporary server failures, connection drops, etc. Below is what my code looks like.
Now, I am having a weird situation over and over: It all starts perfectly. 10 threads pull data concurrently for about 10 minutes. After that time I start getting WebException with timeouts right after I call the GetResponse method of my request object. Taking a break (getting a thread to sleep) doesn't help. It only helps when I stop the application and start it over until the next 10 minutes pass and the problem comes back again.
What I tried already and nothing has helped:
- to close/dispose the response object explicitly and via the "using" statement
- to call request.Abort everywhere it could have helped
- to manipulate timeouts at ServicePointManager/ServicePoint and WebRequest level (extend / shorten the timeout interval)
- to manipulate the KeepAlive property
- to call to CloseConnectionGroup
- to manipulate the number the threads that run simultaneously
Nothing helps! So it seems like it's a bug or at least very poorly documented behavior. I've seen a lot of question regarding this in Google and on Stackoverflow, but non of them is fully answered. Basically people suggest one of the things from the list above. I tried all of them.
public TResource DownloadResource(Uri uri)
{
for (var resourceReadingAttempt = 0; resourceReadingAttempt <= MaxTries; resourceReadingAttempt++)
{
var request = (HttpWebRequest)WebRequest.Create(uri);
HttpWebResponse response = null;
for (var downloadAttempt = 0; downloadAttempt <= MaxTries; downloadAttempt++)
{
if (downloadAttempt > 0)
{
var sleepFor = TimeSpan.FromSeconds(4 << downloadAttempt) + TimeSpan.FromMilliseconds(new Random(DateTime.Now.Millisecond).Next(1000));
Trace.WriteLine("Retry #" + downloadAttempt + " in " + sleepFor + ".");
Thread.Sleep(sleepFor);
}
Trace.WriteLine("Trying to get a resource by URL: " + uri);
var watch = Stopwatch.StartNew();
try
{
response = (HttpWebResponse)request.GetResponse();
break;
}
catch (WebException exception)
{
request.Abort();
Trace.WriteLine("Failed to get a resource by the URL: " + uri + " after " + watch.Elapsed + ". " + exception.Message);
if (exception.Status == WebExceptionStatus.Timeout)
{
//Trace.WriteLine("Closing " + request.ServicePoint.CurrentConnections + " current connections.");
//request.ServicePoint.CloseConnectionGroup(request.ConnectionGroupName);
//request.Abort();
continue;
}
else
{
using (var failure = exception.Response as HttpWebResponse)
{
Int32 code;
try { code = failure != null ? (Int32)failure.StatusCode : 500; }
catch { code = 500; }
if (code >= 500 && code < 600)
{
if (failure != null) failure.Close();
continue;
}
else
{
Trace.TraceError(exception.ToString());
throw;
}
}
}
}
}
if (response == null) throw new ApplicationException("Unable to get a resource from URL \"" + uri + "\".");
try
{
// response disposal is required to eliminate problems with timeouts
// more about the problem: http://stackoverflow.com/questions/5827030/httpwebrequest-times-out-on-second-call
// http://social.msdn.microsoft.com/Forums/en/netfxnetcom/thread/a2014f3d-122b-4cd6-a886-d619d7e3140e
TResource resource;
using (var stream = response.GetResponseStream())
{
try
{
resource = this.reader.ReadFromStream(stream);
}
catch (IOException exception)
{
Trace.TraceError("Unable to read the resource stream: " + exception.ToString());
continue;
}
}
return resource;
}
finally
{
// recycle as much as you can
if (response != null)
{
response.Close();
(response as IDisposable).Dispose();
response = null;
}
if (request != null)
{
//Trace.WriteLine("closing connection group: " + request.ConnectionGroupName);
//request.ServicePoint.CloseConnectionGroup(request.ConnectionGroupName);
request.Abort();
request = null;
}
}
}
throw new ApplicationException("Resource was not able to be acquired after several attempts.");
}