Date: Fri, 31 Oct 2008 16:48:25 +0100 From: Lorenzo Perone <lopez.on.the.lists@yellowspace.net> To: Arnaud Houdelette <arnaud.houdelette@tzim.net> Cc: Louis Kowolowski <louisk@cryptomonkeys.com>, freebsd-stable@freebsd.org Subject: Re: ZFS Message-ID: <EF2EDA0A-E059-4BB2-919F-369CE591EC7A@yellowspace.net> In-Reply-To: <490B0F1D.2050606@tzim.net> References: <FFF7941F7B184445881228ABAD4494B34E7345@intsika.ct.esn.org.za> <200810220838.45900.fjwcash@gmail.com> <43E87CCF-6D36-4F82-BF54-7B705CB1EFB5@yellowspace.net> <1EA86BC6-349F-48DB-A77C-A4D8E00C55B5@cryptomonkeys.com> <490B0F1D.2050606@tzim.net>
next in thread | previous in thread | raw e-mail | index | archive | help
> I use ZFS since 7.0-RELEASE. I'm currently using latest stable.
> Ok the load is not as a production one, as the box is used as a home
> server (NAS), but the hardware is limited too (only 512MB of RAM,
> mono-core A64 3200+, motherborad integrated sata controler).
> I tried to stress the filesystem a bit with multiple simultaneous
> rsyncs. No glitches. The only failures was when swap was on a zvol
> instead of the system drive. Even with more ram, it regularely ended
> in panics or deadlocks (most of the time, deadlocks) under "high"
> load.
>
> Not sure of anything here, but you might want to try with non-zfs
> swap - on another drive(s) or dedicated slices ?
Yep, I think I'm going to use a separate slice
for the pool, mounting into the respective jails
only the needed filesystems:
mypool/mail into /jails/mail/maildataroot
mypool/db into /jails/web/mysql-bup-slave
(or sort of)
and then use frequent snapshots for mypool/mail
(even hourly or so), and for the database,
a few times per day mysql-backup-slave.sh stop,
zfs snapshot mypool/db, mysql-backup-slave.sh start..
the mysql slave snapshotting is really a goodness
which I've used on a SunOS with zfs and really
rocks. So I never shutdown the master, the slave
goes down only a few seconds, and the database
filesystem is consistent and synced.
In the current case, I think it is not only a feature
but also a must: _If_ the host deadlocks and
mysql fails to sync, at least I have a working
snapshot of the data. I wouldn't put the master itself
on zfs for now, but if all goes well for a while,
why not.
BTW: while sync does not work anymore in a deadlock situation,
I've seen that fsync mostly still does.
So something like find /var/db/mysql -type f -exec fsync {} \;
can save your files if the db is running on UFS..
Thanx & Regards!
Lorenzo
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?EF2EDA0A-E059-4BB2-919F-369CE591EC7A>