Hi, there are three possible solutions to this problem. 1) Use "reload" (i.e. graceful restart) instead of "restart" in the logrotate script. Unfortunately, there have been problems with graceful restart in the past, and there are some upstream bugreports about it. 2) Use "graceful-stop" instead of "stop" during restart. This would give the apache children more time to shut down. However, they still would have to close the listening socket in the 10 seconds delay, which might not happen in all cases. 3) Increase the sleep time from 10 to 15 seconds. This might help because apache waits 9 seconds before SIGKILLing stuck children, which leaves only 10-9=1 second to do the killing (which might not be enough on busy servers). But 15-9=6 seconds is much larger than 1 second. Solution 3) is the least likely to break other things, so it would probably be the best solution for Debian etch 4.0r1. It would be a great help if some of you could test whether increasing the timeout helps. Solution 2) has been uploaded to Debian unstable in version 2.2.3-5. Feedback is is very welcome there, too. Note: if you have unstable's 2.2.3-5 and want to try 3), you also have to change "graceful-stop" to "stop" in /etc/init.d/apache2 . Thanks for your help. Stefan
Attachment:
signature.asc
Description: This is a digitally signed message part.