ZFS Wins again - complete power supply failure

P

packetboy

Limp Gawd
Joined
Aug 2, 2009
Messages
288
Had the displeasure of having the OCZ power supply in one of my ZFS servers completely fail yesterday afternoon....especially since I have iSCSI shares with writeback cache enabled (until I get my new ZIL drive installed later this week).

Quick run to Frys this morning for a new power supply and had everything backup by noon...zero data loss...no array problems what-so-ever.
 
yr damn lucky it didnt fry anything
and only in that case if smth would survive we could all hail zfs as saviour ;)
 
packetboy said:
Had the displeasure of having the OCZ power supply in one of my ZFS servers completely fail yesterday afternoon....especially since I have iSCSI shares with writeback cache enabled (until I get my new ZIL drive installed later this week).

Quick run to Frys this morning for a new power supply and had everything backup by noon...zero data loss...no array problems what-so-ever.
Click to expand...

BTW, I have had the same result using mdadm at work for the last 12+ years and now ~50TB spread out over 5 or so servers. Never lost a single byte on a power outage or power supply failure. Although I do have tape backups if anything went wrong..

what does this have to do with ZFS?
Click to expand...

ZFS has guarantees on not corrupting data on loss of power.
 
I second that my gf will just unplug the damn thing when I am not at home not realizing that I will be working on it remote(nor care) and I have never had a problem with dataloss. What platform are you running ZFS on? I am using nas4free.org on a dl380s(12 drive SAS/SATA system) running ZFSv28 and love it.
 
packetboy said:
Point was that even with sudden power loss ZFS integrity was maintained.
Click to expand...

i have had power lost to our office cause of a bad electrition killing our UPS connection, on windows servers, and not have data loss.....

i would put it more down to the fact that at that second, there happened to be nothing being written to the drive....
 
Last edited:
danswartz said:
I believe if you have no separate ZIL device and have disabled sync, and enabled writeback cache on the zvol, the client is in fact vulnerable to data corruption. Interesting article here:

http://permalink.gmane.org/gmane.os.solaris.opensolaris.zfs/43399

It seems the only way to really be safe here is enable sync, or disable writeback cache per LUN or get a fast SSD for ZIL.
Click to expand...
I heard that if you loose the ZIL device, then you might corrupt the zpool only if you are using an old ZFS version.

With the latest ZFS version, if you loose the ZIL, then you only loose the latest writes. Nothing else will happen to your zpool. No corruption, no anything. And this is the reason you should mirror your ZIL device:
http://en.wikipedia.org/wiki/ZFS#ZFS_cache:_L2ARC.2C_ZIL

Thus, if you use ZIL, you should use a new ZFS version, otherwise you might loose your whole zpool if you loose your ZIL device.
 
You misunderstand. I'm not talking about losing the ZIL - that referred to pulling or killing a non-mirrored ZIL device under older versions of the code. The issue here is not ZFS corruption. Remember: he is serving up LUNs via iSCSI, so whatever filesystem the guest has is what is exposed to possible corruption, ZFS does not enter into it.
 
You must log in or register to reply here.
Back
Top