Date: Sun, 21 Nov 2010 23:28:47 +0000 From: Paul B Mahol <onemda@gmail.com> To: Ivan Voras <ivoras@freebsd.org> Cc: freebsd-current@freebsd.org Subject: Re: tmpfs out of space (ZFS related?) Message-ID: <AANLkTi=aUKiJvLToM6PX0Nsc-tEtNZdzHgWAsJNVr-2T@mail.gmail.com> In-Reply-To: <icc77q$mo$1@dough.gmane.org> References: <icc2ih$e5t$1@dough.gmane.org> <AANLkTi=Aw3KQqeDy1Bxn7tJrsvOn-D=y-ChCOpVu4E8N@mail.gmail.com> <20101121214720.356984cc@core.draftnet> <AANLkTimaKY0t4mGtsncSC9dV5SkUdfE%2BMWY5=4F=Xi7X@mail.gmail.com> <AANLkTin1QBZF6KnryEe13-rAk%2BBJOw6uUbyyWq1ef3Co@mail.gmail.com> <icc77q$mo$1@dough.gmane.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, Nov 21, 2010 at 10:42 PM, Ivan Voras <ivoras@freebsd.org> wrote: > On 11/21/10 23:19, Eir Nym wrote: >> >> On 22 November 2010 01:00, Eir Nym<eirnym@gmail.com> =A0wrote: >>> >>> On 22 November 2010 00:47, Bruce Cran<bruce@cran.org.uk> =A0wrote: >>>> >>>> On Mon, 22 Nov 2010 00:43:03 +0300 >>>> Eir Nym<eirnym@gmail.com> =A0wrote: >>>> >>>>> Which type of MFS do you use? I think you shold use "swap-backed" for >>>>> your /tmp, not "malloc-based". Last type is only for in-kernel file >>>>> system. >>>> >>>> It's tmpfs(5), not md(4). >>>> >>> >> >> I wonder if you can create FIFO and sockets with md(4). md(4) is >> geometry, not filesystem. > > md is a GEOM class which presents several types of devices backed by virt= ual > memory. You can create file systems on top of those and mount them. > > The downside is that each of those layers has its own copy of the origina= l > data (and them some). > > tmpfs OTOH is a "real" memory file system, without the need for a backing > device, and it is more efficient with respect to data copies (not perfect= - > it still AFAIK has to maintain something like 2 copies). AFAIK, if this was ever correct it is not any more. Data copies can mean anything. I use >1.5 GB file in tmpfs on 2GB RAM and 0B SWAP without problems. >> tmpfs(5) doesn't support this file types as I can see in FreeBSD source >> code. > > Maybe it is handled by default code, but they are supported. I have used > tmpfs extensively for the last few years for mostly static servers (i.e. > /tmp contains some sockets, a small number of short-lived temporary files= , > etc) - and it so far worked very well. This is actually the first time it > failed me in some way. It can be anything: overflow, race, memory corruption ... Find way to reproduce it 100%.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?AANLkTi=aUKiJvLToM6PX0Nsc-tEtNZdzHgWAsJNVr-2T>