Re: curious crashes when under memory pressure

On 1/4/2025 16:11, Craig Leres wrote:
> On 1/4/25 10:40, Peter 'PMc' Much wrote:
>> There might be a corresponding message in the system log for the
>> "Killed", that would at least tell us the signal, maybe more.
>
> I wish. One minute before I see:
>
>     Jan  4 01:52:10 zinc.ee.lbl.gov kernel: : pid 21007 (conftest), 
> jid 3845, uid 0: exited on signal 11 (core dumped)
>
> but I see a lot of those and I believe they're normal and autoconf 
> related.
>
> On 1/4/25 10:47, Miroslav Lachman wrote:
> > Redundancy in case of disk failure. I've seen many disk failures over
> > the years, and when you have a swap concatenated from partitions on 2
> > disks, the whole system mirrored on 2 disks, and 1 disk fails, the
> > system crashes. That's why we've always used swap over gmirrored
> > partitions. Then we can pull out one disk when system is running and
> > replace it with a new disk.
>
> This. And in my case the system is located in a data center that I do 
> not have casual access to so I'm happy to make the tradeoff in favor 
> of system resilience.
>
>         Craig
>
That's a reasonable justification for mirroring the swap (despite the 
fact that you are then doing two I/Os on a write to it instead of one, 
albeit on different devices) but then I'd still get it out of ZFS' 
domain and run the swap to a bare GPT partition under gmirror instead 
simply because ZFS can demand a RAM allocation to do things and swap 
space is by definition something used under significant memory pressure 
which implies you can need it within ZFS and not be able to get it.

-- 
Karl Denninger
karl@denninger.net
/The Market Ticker/
/[S/MIME encrypted email preferred]/