tmpfs out of space (ZFS related?)
olivier at gid0.org
Mon Nov 22 09:27:53 UTC 2010
2010/11/22 Guido Falsi <mad at madpilot.net>:
> On Mon, Nov 22, 2010 at 12:34:53AM +0100, Ivan Voras wrote:
>> On 11/22/10 00:28, Paul B Mahol wrote:
>> >Find way to reproduce it 100%.
>> The way I described it can be used to reproduce the problem 100%.
>> Detailed instructions for a test case would be something like this:
>> 0) configure a system with tmpfs for /tmp
>> 1) install PostgreSQL 9.0, use a ZFS file system for the database
>> 2) install and use pgbench to initialize a database as large as
>> physical memory (i.e. it will blow away all caches)
>> 3) run pgbench with -c of around 10 or so, let it run for 5 minutes or so
>> 4) stop PostgreSQL, observe reported free memory statistics etc.
> I have seen the same issue, with tmpfs reporting no space.
> I can see it happen regularly when I have big virtualbox machines
> running which take most of the physical mem.
I have the same problem :
FreeBSD 8-STABLE amd64
ZFS v15 as root filesystem
/tmp is a tmpfs
2 GB RAM
1.5GB ARC for ZFS
When compiling "big" ports (like www/firefox), ARC fills wired memory,
and /tmp show 0B available. make fails because it uses /tmp.
If I lower ARC size to 1GB, the same port has no problems to compile.
> Here I have seen it on an 8GB machine with al ZFS when running 3-4 VMs
> taking 1 or 2 GBs each,(cumulatively taking around 7).
> I imagine this is even easier to trigger on machine with less RAM.
> Guido Falsi <mad at madpilot.net>
> freebsd-current at freebsd.org mailing list
> To unsubscribe, send any mail to "freebsd-current-unsubscribe at freebsd.org"
Olivier Smedts _
ASCII ribbon campaign ( )
e-mail: olivier at gid0.org - against HTML email & vCards X
www: http://www.gid0.org - against proprietary attachments / \
"Il y a seulement 10 sortes de gens dans le monde :
ceux qui comprennent le binaire,
et ceux qui ne le comprennent pas."
More information about the freebsd-current