On Fri, Feb 8, 2013 at 10:28 AM, Michael Zender <michael.zen...@mos-tangram.com> wrote:
> I finally solved my problem and wanted to share my solution with you. Thanks for letting us know. > It turned out, that Kaspersky Endpoint Security 8 and its Web-Anti-Virus > feature in particular were causing this problem to show up. We solved it > by defining a rule that excludes our subversion servers from the > Web-Anti-Virus service. The Windows XP still had Kaspersky 6 installed > which does not have the Web-Anti-Virus feature. I was thinking on this overnight, and believe it or not I was going to propose you look in this general direction. > I still don't know what exactly the problem is because in my opinion, > the anti virus software should act in a completely transparent manner > but anyways, it's working now, so I don't bother any more! If you look at the FAQ and remove some of the specifics, more generally what it is saying is that while the client is doing some work the connection to the server is closed in a manner that the client is not expecting it. So the error manifests to the client as a chunk delimiter error when the data it is reading disappears. The FAQ describes one scenario that caused this, the server ending the connection. These Windows anti-virus solutions operate at a low level so they can intercept and monitor your TCP/IP traffic. I would guess that either Subversion's pattern of HTTP requests looked unusual or perhaps even the content in one of your files. My guess is that when these tools sense a problem they do not try to be graceful about it. It probably just kills the connection. After all if it were a virus or trojan horse on your computer it does not want to make it easier for the malicious code to recover. So most likely when it senses a problem it closed the connection and that manifested itself the same as the server timeout. One thing that might be helpful is to look into what kind of logging the tool provides. It would be nice if they log some forensic data about what caused them to do this. Maybe that information can go back to Kapersky to make the tool not do this. Or maybe it is just a bug in their tool where they cannot handle all of the requests and how fast they are being made. I suspect a SVN client drives HTTP traffic a lot differently than a typical web browser loading a page does. > Thanks again to Mark for his reply, it definitely made me investigate in > the right direction. You are welcome and thanks for sharing the information back. Do you have any suggestions on how this FAQ could be improved to add this information? -- Thanks Mark Phippard http://markphip.blogspot.com/
