Hi. I'm haunted by a weird bug. Some of my servers (IBM x3250) hang periodically. And this is always saturday morning. Different servers in different cities, all with zfs and one gig of RAM. And yeah, it's periodic weekly. I can say more - it's repeatable. 25 minutes ago I typed 'periodic weekly', and 5 minutes ago I lost this machine from the network (even stopped answering to ICMP). This can be solved by any of two methods - either increase RAM, or turning off periodic weekly. loader.conf (doesn't help): zfs_load="YES" vfs.root.mountfrom="zfs:zfsroot" ng_iface_load="YES" ng_ether_load="YES" vm.kmem_size="330M" vm.kmem_size_max="330M" vfs.zfs.arc_max="30M" This is 8.2-RELEASE/amd64. The most important question - what to report and how to report this ? Thanks. Eugene.
On 02/27/2012 06:07 AM, Eugene M. Zheganin wrote:> Hi. > > I'm haunted by a weird bug. > Some of my servers (IBM x3250) hang periodically. And this is always > saturday morning. Different servers in different cities, all with zfs > and one gig of RAM. And yeah, it's periodic weekly. I can say more - > it's repeatable. 25 minutes ago I typed 'periodic weekly', and 5 > minutes ago I lost this machine from the network (even stopped > answering to ICMP). This can be solved by any of two methods - either > increase RAM, or turning off periodic weekly. > > loader.conf (doesn't help): > > zfs_load="YES" > vfs.root.mountfrom="zfs:zfsroot" > ng_iface_load="YES" > ng_ether_load="YES" > vm.kmem_size="330M" > vm.kmem_size_max="330M" > vfs.zfs.arc_max="30M" > > This is 8.2-RELEASE/amd64.8.2-RELEASE is highly unstable with ZFS in my opinion. For example, my system with 48 GB of RAM would hang or crash for no apparent reason in random intervals. Upgrading in September fixed most of it, except 1 random hang possibly related to NFS, and a hang when renaming snapshots with zvols [new problem after 8.2-RELEASE, PR 161968]. The renaming snapshot with zvols hang seems fixed since update again in February. Still unresolved though is that restarting nfsd makes nfsd hang until reboot. I don't know if 8.2-RELEASE had that problem. You should consider an update with csup. Make sure you test and monitor after the upgrade, in case some other funny issues come up, like my restarting nfsd issue. But of course I chose that over random hangs and panics. And one word of advice: If you want to upgrade your pools to v28, I think you should consider recreating your pools as v28 rather than upgrading. There are some side effects to upgrading, such as logs that can't be removed.> > The most important question - what to report and how to report this ? > > Thanks. > Eugene. > _______________________________________________ > freebsd-stable@freebsd.org mailing list > http://lists.freebsd.org/mailman/listinfo/freebsd-stable > To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"-- -------------------------------------------- Peter Maloney Brockmann Consult Max-Planck-Str. 2 21502 Geesthacht Germany Tel: +49 4152 889 300 Fax: +49 4152 889 333 E-mail: peter.maloney@brockmann-consult.de Internet: http://www.brockmann-consult.de --------------------------------------------
On 27/02/2012 05:07, Eugene M. Zheganin wrote:> Hi. > > I'm haunted by a weird bug. > Some of my servers (IBM x3250) hang periodically. And this is always > saturday morning. Different servers in different cities, all with zfs > and one gig of RAM. And yeah, it's periodic weekly. I can say more - > it's repeatable. 25 minutes ago I typed 'periodic weekly', and 5 > minutes ago I lost this machine from the network (even stopped answering > to ICMP). This can be solved by any of two methods - either increase > RAM, or turning off periodic weekly.You could try to narrow it down to one specific script. My first guess is that 310.locate brings the machine down as it traverses the whole tree.> > loader.conf (doesn't help): > > zfs_load="YES" > vfs.root.mountfrom="zfs:zfsroot" > ng_iface_load="YES" > ng_ether_load="YES" > vm.kmem_size="330M" > vm.kmem_size_max="330M" > vfs.zfs.arc_max="30M" > > This is 8.2-RELEASE/amd64. > > The most important question - what to report and how to report this ? > > Thanks. > Eugene. > _______________________________________________ > freebsd-stable@freebsd.org mailing list > http://lists.freebsd.org/mailman/listinfo/freebsd-stable > To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org" >
Hi-- On Feb 26, 2012, at 9:07 PM, Eugene M. Zheganin wrote: [ ... ]> all with zfs and one gig of RAM.This isn't a sensible combination; I wouldn't try to run ZFS on anything less than 4GB... Regards, -- -Chuck
On Tue, February 28, 2012 01:10, Eugene M. Zheganin wrote:> Hi. > > On 28.02.2012 01:02, Nenhum_de_Nos wrote: >> regardless of the pool size ? >> >> I was planning on making an atom board a file server for my home, and I have two options: soekris >> net6501 2GB RAM and intel board powered by the 330 atom (says 2GB limited as well). My plansare to use from 4 up to 8 disks, and they should be 2TB at least.>> >> As its for home use, some p2p software and mostly music listening and sometimes movie streaming. >> >> should 2GB be that bad, that I should drop it and use UFS instead ? >> >> I may run any version of FreeBSD on it, was planning on 9-STABLE or 9.1. >> > In the same time I have a couple of hosts successfully running zfs on 768 Megs and on 1 Gig ofRAM. Both i386.> And they aren't affected by the periodic weekly for some reason. And they are used only asfileservers.> > So when I see all these advices to add a gazillion gigabytes of RAM to use zfs - I don't see theconnection. Eugene, what's the pool size ? I'd like to use one using 4x 2TB disks (may be 3TB). and other using random disks that may evolve to that same amount. matheus -- We will call you Cygnus, The God of balance you shall be A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing? http://en.wikipedia.org/wiki/Posting_style
On Feb 26, 2012, at 11:07 PM, Eugene M. Zheganin wrote:> I'm haunted by a weird bug. > Some of my servers (IBM x3250) hang periodically. And this is always saturday morning. Different servers in different cities, all with zfs and one gig of RAM. And yeah, it's periodic weekly. I can say more - it's repeatable. 25 minutes ago I typed 'periodic weekly', and 5 minutes ago I lost this machine from the network (even stopped answering to ICMP). This can be solved by any of two methods - either increase RAM, or turning off periodic weekly.What is the output from 'swapinfo'? -- Jason Wohlford <jason@wohlford.org> <http://wohlford.org/> @wohlford
on 27/02/2012 07:07 Eugene M. Zheganin said the following:> vfs.zfs.arc_max="30M"I am not much of an expert in tuning ZFS, and for low-end systems in particular... But it seems to me that with this setting you are actually stress ("torture") testing ZFS rather than using it. -- Andriy Gapon