Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 21 Sep 2018 11:17:43 -0400
From:      Josh Gitlin <jgitlin@goboomtown.com>
To:        freebsd-fs@freebsd.org
Subject:   Re: Troubleshooting kernel panic with zfs
Message-ID:  <373102CE-B53A-4547-8665-DF5EC74C23DD@goboomtown.com>
In-Reply-To: <c4d7e51f-09c9-c50c-075d-8eff65e64abe@sentex.net>
References:  <D54225AD-CC96-45E7-A203-D2C52E984963@goboomtown.com> <c4d7e51f-09c9-c50c-075d-8eff65e64abe@sentex.net>

next in thread | previous in thread | raw e-mail | index | archive | help
Thanks for the help Mike! In reviewing our settings I also found that =
we're setting vm.kmem_size to 1.5x physical RAM installed, based on a =
tuning guide someone found. I personally believe that might be misguided =
and may be related to the panic we saw, as this causes zfs-stats -a to =
show the kernel memory map being larger than system RAM:

Kernel Memory Map:				24.00	GiB
	Size:				21.72%	5.21	GiB
	Free:				78.28%	18.79	GiB

I'll try tuning these settings down and see if the problem happens =
again. Appreciate the support!

--
 <http://www.goboomtown.com/>=09
Josh Gitlin
Senior DevOps Engineer
(415) 690-1610 x155

Stay up to date and join the conversation in Relay =
<http://relay.goboomtown.com/>.

> On Sep 21, 2018, at 9:04 AM, Mike Tancsa <mike@sentex.net> wrote:
>=20
> On 9/20/2018 8:27 PM, Josh Gitlin wrote:
>>=20
>> My hunch is that, given this was inside kmem_malloc, we were unable =
to allocate memory for a zio_write_compress call (the pool does have ZFS =
compression on) and hence this is a tuning issue and not a bug... but I =
am looking for confirmation and/or suggested changes/troubleshooting =
steps. The ZFS tuning configuration has been stable for years, to it may =
be a change in behavior or traffic... If this looks like it might be a =
bug, I will be able to get more information from a minidump if it =
reoccurs and can follow up on this thread.
>=20
> Given all the issues with memory pressure perhaps ARC is eating too
> much, I would try artificially limiting it so it does not eat up too
> much RAM.  We had a similar purposed server but with 32G RAM and I =
found
> one day the system was randomly shooting processes to reclaim RAM
>=20
> eg
> pid xxx (yyy), uid zzzz, was killed: out of swap space
>=20
>=20
> Perhaps
>=20
> sysctl -w vfs.zfs.arc_max=3D6000000000
>=20
> eg.
> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D229764
>=20
> =
https://lists.freebsd.org/pipermail/freebsd-questions/2018-July/282261.htm=
l
>=20
> 	---Mike
>=20
>=20
>=20
>=20
> --=20
> -------------------
> Mike Tancsa, tel +1 519 651 3400 x203
> Sentex Communications, mike@sentex.net
> Providing Internet services since 1994 www.sentex.net
> Cambridge, Ontario Canada




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?373102CE-B53A-4547-8665-DF5EC74C23DD>