DragonFly bugs List (threaded) for 2005-02
Re: vnlur_proc lockups under high disk I/O
Matthew Dillon wrote:
:I've attached the dmesg output. I'm running memory and disk diagnostics
:as it's strange that its only just started. (I haven't run a buildworld
:in a week but only noticed the problem yesterday.)
:Extracting a large tar file while the CPU is under load nearly always
:triggers the problem. It's as if disk IO operations take longer and
:longer until they eventually time out locking the machine. You can watch
:the disk activity led go from constant to briefly flashing at increasing
:Buildworlds also gradually slow to a crawl but so far haven't resulted
:in the vnlru_proc errors.
:The first time it happened I was running "make -j 4 buildworld" on the
:console and extracting the pkgsrc tarball to /usr/pkgsrc via a SSH session.
What is likely happening is that the recycler is trying to flush out vnodes
related to things other then the hard drive, and one of those things is
causing it to stall.
Observe your other mounts carefully. What other mounts do you have?
I have a semi-production machine running stable from a few months ago.
The vnode related sysctl values on this machine are;
kern.maxvnodes = 36053
kern.minvnodes = 9008
However on the machine I'm having problems with they are; (current stable)
kern.maxvnodes = 16754
kern.minvnodes = 4188
I've added the previous settings to /etc/sysctl.conf and the problems
have not reoccurred. I can now run "cd /usr/pkgsrc && cvs -q update -dP"
without problems, previously this would result in a lockup without failure.
I only have standard mounts and am only working within /usr.
Filesystem 1K-blocks Used Avail Capacity Mounted on
/dev/ad2s1a 254063 47193 186545 20% /
/dev/ad2s1d 254063 307 233431 0% /var
/dev/ad2s1e 254063 8 233730 0% /tmp
/dev/ad2s1f 36606155 346959 33330704 1% /usr
procfs 4 4 0 100% /proc
Do you still want a kernel core?