Hey guys, A status update on the crashes. I have identified what I think are 3 different problems.
1.) There's a bug in the replay system due to a flaw in libcurl using a signal to handle DNS timeout. You can avoid this bug by using IP addresses in your replay config, rather than DNS names. We will have a software workaround in the next update or so that essentially does this same thing automatically. 2.) There's a random memory scribble. It will manifest itself as "double free" or "memory corruption" crash, depending on your OS. Some have theorized than this is due to the Dr. G weapons. We cannot confirm this. 3.) There is a hang. From what information I have gathered, the last thing in the log is something along the lines of "PreMinidumpCallback: updating dump comment." In other words, it is hanging while attempting to report the crash. This is particularly disastrous because it not only will it interfere with auto-restart scripts (unless you have some sort of watchdog), but it prevents the crash report from being generated and submitted, which of course would help us fix it. A random memory scribble can cause all sorts of behaviour, so it's possible that #2 is the real bug, and #3 just a side effect that sometimes attends the main bug. We have not been able to reproduce any of these issues internally, and we have had several playtests. (We, the actual developers, not a separate QA department and not a group of interns, playtest the game every day, on Windows and Linux servers.) However, our dedicated servers have experienced the hang. It has been very difficult to track down and fix these crashes because we seem to have several regressed all at once, and at least one of the problems is interfering with the normal reporting mechanism. If anyone is able to save a dump file (they usually go to /tmp/dumps), I would be great if you could post them in some webspace and post a URL where they may be downloaded. Or, if your console log shows that it was uploaded, please post the report ID. The output will look something like this: PreMinidumpCallback: updating dump comment Uploading dump (in-process) [proxy ''] /tmp/dumps/crash_20110723191817_1.dmp success = yes response: CrashID=bp-445d6055-e9e7-420a-93b8-688a92110723 Grabs of GDB stack traces, etc with raw addresses, are not totally useless, but they are definitely much less useful. Even with symbols, a stack trace does not have as much data as a dump has. So if we could get some actual dumps, that would be really great. These crashes continue to be our top priority. - Fletch _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
