Re: You're right, I was completely wrong and the freezes I'm seeing are... who even knows what

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

Ellie <el%horse64.org@localhost> writes:

> I'm on a pretty normal x86_64 PC yes. I'm on NetBSD 11.0.

That should be  new enough for zfs locking fixes.

> One of the crashes was very interesting because it happened right as
> memtester seems to have tried to mlock() 8GB of memory:

In order to figure out what's wrong, you're going to need to keep
rigorous notes and modify one thing at a time, assuming you can get to a
configuration that is ok.  And, reading about ulimit setting so you can
provoke a system with and without zfs equally hard in terms of memory
deficit.

> Is this some sort of memory management issue then? I don't know why
> this would happen, I'm just quite confused...

The code to allocate memory, and to free memory, including by paging,
has complicated locking.  It's possible there's a bug.  In general
that's how the world is, both in terms of "code has locking" and
"locking is very hard to get 100% right".

However, calling mlock on even half of RAM is in my view abusive.   One
of the main mechanisms for freeing memory is paging.    If you want to
persist and try to debug the situation and maybe find a bug, I wish you
luck, but I really can't do it remotely via email.