On Wed, 5 Aug 1998, Dax Kelson wrote:
> We discovered 3 deadlock situtations in the Solaris 2.5.x/2.6 kernel
> (which results in unkillable processes). They only occur on a busy box.
*Bang!* So it *was* a Solaris bug we were fighting recently? Box is
definitely "busy", Ultra Enterprise 2/2170, running 2.5.1 - Squid would hang,
machine was 99.8% idle, 0.2% in the kernel; Squid process couldn't be killed,
nor could any process (eg., "sync") that dared to access the cache disks (ie.,
shutdown didn't work, you had to be at the console to manually reboot the
machine).
> 2 of the 3 are fixed in the latest kernel patch, the 3rd is fixed in a
What rev level of the kernel jumbo (103640)?
> Since the last binary fix was installed, we have an uptime of 5 months.
Groovy, thanks..
> > machine doesn't even work (I guess solaris needs all processes to be
> > dead before shutting down, so since it can't kill it, it just sits there
> > looking dumb)
For us, the processes run during shutdown (eg., "sync" or "umount") would
touch the cache disk, lock up and eventually the shutdown procedure wouldn't
finish.
Cheers..
dave
Received on Wed Aug 05 1998 - 16:55:39 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:41:27 MST