FreeBSD 10.1 crashing under load

[Christian Kujau]

On Mon, 16 Feb 2015 at 11:51, Fabian Keil wrote:
> Using a kernel with WITNESS and INVARIANTS sometimes results in
> the problem being caught earlier.

Hm, I had hoped not to have to recompile the kernel for this, but maybe 
that's the way to go. I'll look into that, thanks.

> IIRC it's not officially supported but you can dump on the device
> below the geli layer:
> 
> fk at r500 ~ $swapinfo 
> Device          1K-blocks     Used    Avail Capacity
> /dev/ada0s1b.eli   2097152        0  2097152     0%
> fk at r500 ~ $dumpon -l
> ada0s1b

(FreeBSD n00b here): indeed, even with dumpdev="AUTO", "dumpon -l" 
reported /dev/null - I've set this manually to my encrypted swap device 
now and wait for the next boot. The last involuntary reboot was just a few 
hours ago, the last known state was:

> Mem: 70M Active, 663M Inact, 208M Wired, 4508K Cache, 87M Buf, 32M Free
> Swap: 1024M Total, 88M Used, 936M Free, 8% Inuse

> Maybe you need to play with vm.overcommit for details see tuning(7).

vm.overcommit was set to 0 by default and I would've assumed that this is 
sufficient:

> Setting bit 0 of the vm.overcommit sysctl causes the virtual memory 
> system to return failure to the process when allocation of memory causes
> vm.swap_reserved to exceed vm.swap_total.

But instead of one process crashing, the whole machine crashes. I've set 
it to 2 now and will see if it helps anything.

Thanks for your input,
Christian.
-- 
BOFH excuse #78:

Yes, yes, its called a design limitation