ChatGPT to the rescue
I tend to be very critical of AI in general and generative AI in particular. I think that these systems have a tendency to produce “bullshit” instead of real information, and one has to be very careful when using them. So, I was pleasently surprised, when ChatGPT helped me figure out a problem with WSUS that I had been debugging for hours.
Windows Update scans against the WSUS server kept failing on a whole group of clients with error code 0x80240439, which turned out to be caused by an HTTP 413 “request entity too large” error on the server. I found many articles online that described this problem, but only few contained useful information. The most useful ones referred to the Web.config file in the ClientWebService directory, but even increasing all these limits did not help.
At some point, I thought I could ask ChatGPT for a solution, even though I did not expect that it would give me any additional insights. As expected, ChatGPT basically presented me the information that I had already found, but it suggested that I might want to turn on failed request logging in IIS, in order to confirm that this is indeed a server-side problem. I did so, even though I did not expect to get more information from this, and it indeed confirmed what I had already suspected: The server was rejecting the requests because of a request body that was too large. At least, the information showed me that the request contained a lot of InstalledNonLeafUpdateIDs and even more OtherCachedUpdateIDs, and that only 49152 bytes of the (presumably longer) request body where processed.
The fact that this was much less than any of the limits in Web.config, gave the relevant hint: After giving ChatGPT this information and telling me that I had already tried all the suggested solutions, it gave me a list of less-common items to consider. This list included the uploadReadAheadSize, and a link to a discussion where increasing this size had indeed resolved the problem. I tried it, and it also resolved the problem for me. I wrote up the full solution in the Wiki, in case you are interested.
I have to admit that without ChatGPT, I probably wouldn’t have found this solution. I still believe that ChatGPT on its own is not very useful, as indicated by the fact that it gave me a lot of irrelevant information and I still had to carry out a lot of debugging steps on my own. But when used cautiously, taking all the information provided with a grain of salt and checking the original sources, it can sometimes be useful.