2013/3/14 Thomas, Steve <stho...@vocollect.com>: > Hi - > > Running Tomcat 7.0.23 or 7.0.37 (32 or 64-bit) installed as a service (either > via service.bat or the exe installer) on a Windows 7 64-bit OS, we are seeing > an issue where the Windows shutdown kills Tomcat before our webapp shutdown > sequence has time to execute fully. (Specifically, we just want to make sure > our instance of HSQLDB shuts down correctly, otherwise corruption can ensue.) > > Details: > > Initially we were running with 32-bit Tomcat 7.0.23 and saw that our shutdown > sequence was not being logged at all when one of our customers shut down his > laptop. It looked like the process was just being killed. I found a > commons-daemon/procrun bug and corresponding fix that seemed like it should > address the issue, namely > > http://stackoverflow.com/questions/13578196/how-to-gracefully-shutdown-procrun/14150785#14150785 > > https://issues.apache.org/jira/browse/DAEMON-274 > > I subsequently upgraded Tomcat to 7.0.36 (32-bit, zip) to get the updated > commons-daemon (http://tomcat.apache.org/tomcat-7.0-doc/changelog.html), but > to no avail. I thought perhaps the 32-bit Tomcat/64-bit OS might be a > disconnect, so I installed the 64-bit version, but got the same result. In > short, it looks like we're either doing something wrong with our code, or > there's a new wrinkle in the OS-service handshaking, or the bug wasn't fixed > correctly...maybe in that order. > > Details below on how our code manifested the problem as well as other steps > to reproduce. > > Our database shutdown code is located in the destroy() function of a class > implementing org.springframework.beans.factory.DisposableBean. I added a > Thread.sleep(5 min) call to reproduce it on my machine. As long as I shut > down the service through the Services panel on Windows, the shutdown sequence > fully executes (and takes 5 min, as expected). But if I just shut down > Windows, the sequence is interrupted.
1. From the above I would say that is a Windows feature, that it does not wait for service to shutdown properly. I'd look into Windows support and MSDN, whether it is a configuration issue or something that be workaround by proper coding. http://support.microsoft.com/kb/146092/en-us http://msdn.microsoft.com/en-us/library/windows/desktop/dd371756%28v=vs.85%29.aspx 2. On the "Shutdown" tab of the procrun service configuration dialog (Tomcat7w.exe) there is "Timeout" field. I have "0" (sec) there. 3. Does your database need 5min to shutdown? Are you doing something unnecessary (such as compacting it instead of a simple shutdown)? Are you using a lot of "memory" tables that HSQL writes out as SQL, instead of "cached" (on-disk) binary ones? 4. I have several HSQL databases running in-process in Tomcat for my personal needs. They are configured as <Resource closeMethod="close" url="...;shutdown=true" /> in server.xml. They shutdown together with Tomcat (without any additional listeners) and I did not notice any problems. 5. Is HSQL database corruptible by sudden kill? AFAIK it has own journal and backup, and there is nothing wrong with aborting it. The only thing that when it is properly shut down, it looks more pretty with lesser number of files. > (As an aside, I don't expect the shutdown to take anywhere near that long in > practice, but wanted to make sure the problem manifested itself so that I > could address the bug. We are seeing this with a decent customer test > machine, but I can't reproduce it on my machine w/o changes.) > > Thinking it might be Spring, I moved the shutdown delay to a > ServletContextListener. contextDestroyed() method. Same effect. > > I moved the delay again, and reproduced the same problem in a standalone > servlet that overrides HttpServlet.destroy(). I've posted the code at the > link below: > > http://pastebin.com/yYgrQ2sE > > This is the output recorded in the stdout log file for an OS shutdown, and > then a shutdown of the service by hand. We, of course, favor the second. :-) > > 2013-03-14 10:05:40 Commons Daemon procrun stdout initialized > StatusServlet.init() > Entering StatusServlet.destroy() > Simulating long shutdown sequence. > > 2013-03-14 10:12:29 Commons Daemon procrun stdout initialized > StatusServlet.init() > Entering StatusServlet.destroy() > Simulating long shutdown sequence. > Simulation complete--sequence finished. > Exiting StatusServlet.destroy() > > Can we guarantee that Windows won't just kill our Tomcat process and > potentially corrupt our database? That's the question. > > I'd be grateful for some help on this. Thanks for your time and attention. > Best regards, Konstantin Kolinko --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
