Hi Cyril, On Sat, Apr 02, 2011 at 01:33:54AM +0200, Cyril Bonté wrote: > Hi all, > > Le vendredi 1 avril 2011 07:47:22, Willy Tarreau a écrit : > > Hi Charles, > > > > On Thu, Mar 31, 2011 at 02:42:54PM -0500, Charles Duffy wrote: > > > Howdy! I'm having trouble with the haproxy reload command. The below > > > happens on my VPS vendor's builds of Linux 2.6.38 but not their 2.6.35: > > (...) > > > > I'm not aware of any required kernel config option for this. It's scary, it > > sounds like a kernel regression preventing the process from binding to a > > TCP port it was using a few seconds ago. Do you know if you still have > > established connections when you reload ? We could imagine that the kernel > > refuses to bind in this case (which still would be a regression which > > should be reported). > > I've upgraded to the 2.6.38 kernel provided in Ubuntu natty and it's amazing > to see how it's easily reproducible without any condition. Just start haproxy > and reload it. > > I suspect this kernel commit to be responsible of that behaviour: > http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=c191a836a908d1dd6b40c503741f91b914de3348
Good catch, it looks like you're right indeed. I'll build this kernel for one of my machines and test. If this problem is true, it will have more impacts than haproxy alone, as it will also prevent any fork() based service from being restarted (eg: cannot restart sshd from an ssh session). Willy

