Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 13 Dec 2017 11:23:46 +0000
From:      Gary Palmer <gpalmer@freebsd.org>
To:        blubee blubeeme <gurenchan@gmail.com>
Cc:        Mark Millard <markmi@dsl-only.net>, FreeBSD Current <freebsd-current@freebsd.org>
Subject:   Re: get_swap_pager(x) failed
Message-ID:  <20171213112346.GA27816@in-addr.com>
In-Reply-To: <CALM2mEnftYe4P=Gqx1tE=jUYDSQq4Wh-7_TqnLV_DJYie8%2BtBw@mail.gmail.com>
References:  <C19068B6-2578-41B8-9076-97448C2AF5CB@dsl-only.net> <CALM2mEnftYe4P=Gqx1tE=jUYDSQq4Wh-7_TqnLV_DJYie8%2BtBw@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Dec 13, 2017 at 10:34:02AM +0800, blubee blubeeme wrote:
> On Wed, Dec 13, 2017 at 5:53 AM, Mark Millard <markmi@dsl-only.net> wrote:
> 
> > blubee blubeeme gurenchan at gmail.com wrote on
> > Tue Dec 12 15:58:19 UTC 2017 :
> >
> > > On Tue, Dec 12, 2017 at 3:34 PM, blubee blubeeme <gurenchan at gmail.com
> > > >wrote:
> > > > I am seeing tons of these messages while running tail -f
> > /var/log/messages
> > > > ============
> > > > Dec 12 15:11:41 blubee kernel: swap_pager_getswapspace(25): failed
> > > . . .
> > > >  1159 blubee        5  20    0   149M 56876K select  6   1:05   0.00%
> > > > ibus-engine-chewing
> > > >
> > > > ===========
> > > >
> > > > What's with all the swap errors? I am running ZFS and I have 16GB of
> > ram,
> > > > how could I be having swap space errors?
> > > >
> > >
> > > Well I added 4GB of extra swap in /var/tmp/swap0
> > > then added that to my /etc/fstab: md99                    none    swap
> > > sw,file=/var/tmp/swap0,late     0       0
> > >
> > > and those errors went away.
> >
> > I recommend reviewing bugzilla 206048 (title in part
> > "swapfile usage hangs; swap partition works"):
> >
> > https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=206048
> >
> > before using file-system based swap spaces. They have
> > lots of problems with deadlocks. See especially comments
> > #7 and #8 quoting Konstantin Belousov. #8 is just a
> > reference to:
> >
> > https://www.freebsd.org/doc/en_US.ISO8859-1/books/developers-handbook/
> > kerneldebug-deadlocks.html
> >
> > Comment #3 shows a way to test for the problematical
> > behavior.
> >
> > Using swap partitions avoids the issue.
> >
> > ===
> > Mark Millard
> > markmi at dsl-only.net
> >
> >
> Thanks for the info, why would I be getting swap errors like that when I
> have 32GB of ram?
> sysctl hw.physmem
> hw.physmem: 34253692928
> 
> That really doesn't make any sense to me... Is it Chromium eating up 32GB+
> of ram?

Your top output from the first message in the thread gives a hint:

Mem: 3442M Active, 293M Inact, 4901M Laundry, 22G Wired, 1057M Free             
ARC: 18G Total, 1488M MFU, 15G MRU, 4003K Anon, 178M Header, 1171M Other        
     16G Compressed, 23G Uncompressed, 1.41:1 Ratio                             
Swap: 2048M Total, 2028M Used, 20M Free, 99% Inuse, 36K In                      

You have 22GB RAM used by the kernel, and 18GB of that is used by ZFS.
As you can see from the last line, 99% of your small swap space is used
and paging was happening at the time of the snapshot ("36K In").

First I would suggest limiting ARC and seeing if that helps.  Unless you
are doing a lot of fileserver type work or working with lots of files
you want to keep around that ARC size is a bit big.  My desktop with
32GB RAM has ARC limited to 4GB.

Definitely do NOT swap to a file on a filesystem, especially if the
filesystem is ZFS.  That will eventually lead to a deadlock.

An open question would be why ARC is not reducing if the system is
under memory pressure.  It's meant to, but there have been various
bugs in that implementation.

Regards,

Gary



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20171213112346.GA27816>