Den fre 17 maj 2024 kl 00:20 skrev Williams, James P. {Jim} (JSC-CD4)[KBR Wyle Services, LLC] via users <users@subversion.apache.org>:
> > BTW: as Daniel asked: You previously mentioned Subversion 1.14.1, is > > that on the server or on the client? > > I've been testing with SVN 1.14.1 on both the client and server. > Perfect, thanks for confirming. Do you happen to know what version of Serf is used on the client and what version of OpenSSL do you have on the client/server? > > > Regardless, since the issue manifests on the client-side by hangs and > > crashes while waiting for / processing data from the server (hangs > > inside libsvn_ra_serf), I'd suggest also to investigate whether there > > is something in between your client and your server that might be > > interfering. A proxy or reverse proxy perhaps? Or some security > > software on the client that interferes with the network (as some > > antivirus suites do on Windows)? If so, try to bypass it (or disable / > > create an exclude rule), as a diagnostic step to see whether this > > might be the cause. > > I don't think we have anything inserted between client and server, but > will ask those smarter than me if they know of such a thing. > > I do have an update. My system administrator was surprised to see the > server machine configured to support both version 4 and 6 IP addresses. He > thought the latter were turned off. After that change, checkouts were > noticeably faster, and hangs and crashes were noticeably less frequent. > The repo size needed to trigger them appeared to grow some. Though a step > forward, the configuration still isn't useable. It might also point at > further network configuration problems as a cause, though the SVN client > probably shouldn't crash in any case. > Are the client and the server both "close" (networkwise) or do they communicate over a WAN/internet link? Is it possible to - at least temporarily - add a VHost without SSL encryption to see if this makes a difference? That would also make it easier to catch the traffic using Wireshark to see the last packages before a crash. In particular, I'm interested in the content of the packages that trigger the xml parsing error (the call to svn_xml_parse). For the cases when it hangs, does one side say "goodbye" (ie, FIN) or does it seem like package drops (ie, traffic just stops flowing). When it hangs, does it ever release? I would assume that the client would finally time out. Kind regards, Daniel