TCP problems after 124 days of uptime?

[Gleb Smirnoff]

On Thu, May 24, 2007 at 01:52:33PM -0700, Luigi Rizzo wrote:
L> On Thu, May 24, 2007 at 09:59:41PM +0200, Andre Oppermann wrote:
L> > Gleb Smirnoff wrote:
L> > >   Yesterday two of our web servers running 6.2-PRERELEASE experienced
L> > > problems with accepting TCP connections. A small percentage of SYN
L> > > packets was ignored. The packets were seen in tcpdump output, but not
L> > > processed by TCP stack. No accept queue overflows occured, according
L> > > to 'netstat -sp tcp'.
L> > > 
L> > > Failing to find any clue I have rebooted one of them, and after
L> > > reboot is started to work flawlessly. Reboot also helped the second
L> > > one. Both servers were booted at the same time - 124 days ago.
L> > > 
L> > > Any ideas, any similar reports?
L> > 
L> > ticks is a 32 bit signed integer and TCP isn't really aware of it
L> > going negative and rolling over.  This is something I'm working on
L> > fixing in -current.  Haven't analyzed all potential cases there yet.
L> 
L> by chance, is your 'hz' set to 200 ?
L> this might explay something, because 2^31/(86400*200) = 124.275

Nope. hz is default - 1000.

-- 
Totus tuus, Glebius.
GLEBIUS-RIPN GLEB-RIPE