Re: zfs questions

[Greg Troxel <gdt%lexort.com@localhost>]

Takashi YAMAMOTO <yamt9999%gmail.com@localhost> writes:

> On Fri, Feb 20, 2026 at 10:41 PM Greg Troxel <gdt%lexort.com@localhost> wrote:
>>
>> Takashi YAMAMOTO <yamt9999%gmail.com@localhost> writes:
>>
>> > * is anyone using zfs in netbsd seriously? :-)
>>
>> I am using it on my desktop, and a on a dom0 hosting pkg builders.
>>
>> It mostly works, except that one must be careful not to simultaneously
>>
>>   run low on memory
>>   run programs that dirty mmap'd pages (syncthing!)
>
> is it enough to run syncthing from pkgsrc?
> i tried to make it receive 1024 * 1GB files. it just worked fine for me.

It is not enough to just run syncthing (and yes from pkgsrc).

The system needs to be under significant memory presssure.

This happens to me often enough, if I don't avoid it.

  32 GB RAM, 8 cores
  build.sh -j8 or -j16, or pkg_rolling-replace with high MAKE_JOBS
  firefox running
  syncthing receiving a lot
  default maxvnodes

I avoid these lockups by three things

  not keeping syncthing running (only starting syncthing with no builds,
  and soemtimes I shut down firfox)

  maxvnodes 200000 (vs far far higher by default)

  keeping MAKE_JOBS at 2, maybe 4

I do run build.sh with much higher -j values, but pkgsrc builds things
with rust, and apparently the rust developers have never had to use a
computer without a vast ramount of RAM.




I think part of the issue is that a vnode being in the cache might be
enough to keep it sort of in the ARC.  I don't really understand that,
but anecdata is strongly in support of "lower maxvnodes is associated
with reduced frequency of lockups".

I suggest booting up a 4 GB machine in a VM  and stressing that.