How many of you have experienced catastrphic pool failure and why?

[Knowltey]

Maximum space and speed.

You did read I have 2 identical systems right?

Running RAID0 in a server is still completely idiotic.

 

[SLIMaxPower]

I'm open to suggestions if it doesn't involve mirroring or spares

 

[Knowltey]

I'm open to suggestions if it doesn't involve mirroring or spares

RAIDZ1 if you don't want mirroring.

 

[gpsguy]

Yes, and assuming the backup server isn't protected by a UPS, there's a chance a power failure could damage both systems.

You did read I have 2 identical systems right?

 

[SLIMaxPower]

Yes, and assuming the backup server isn't protected by a UPS, there's a chance a power failure could damage both systems.

Second system is only powered on for weekly rsync

 

[jgreco]

It compromises ZFS's self-healing data capability to do what you're doing, and also the likelihood of pool loss increases rather dramatically: The probability that a drive fails in any given month isn't all that comforting a number, and being dependent on two of them simultaneously... much worse.

 

[SLIMaxPower]

I understand the reasoning for ZFS and raidz for data protection, but I am after maximum space.

With the size of HP Microservers 7 or 8 drives is the maximum. I use 7 drives split between 2 pools and the 8th drive for jails. I am currently using 4x3tb raid-0 10.7tb total 76% used, 3x3tb raid-0 8tb total 80% used. 1x2tb jails. Losing one drive minimum from either pool will leave my current storage needs short.

Moving up to 5 or 6tb drives around xmas for both servers.

Have prob had 2 drive failures over 6 years, and rsync r done weekly.

 

[jgreco]

I should probably have clarified that I'm perfectly fine with you compromising the resiliency features, as long as you're aware of the downsides of doing so. There's some irony to it being mentioned in this thread, though...

 

[DaPlumber]

Second system is only powered on for weekly rsync

This matters not diddly squat. When you multiply odds they increase, or without going into the math which would require guess-timating MTBF's etc:

You would dramatically lower your risk of catastrophic data loss by combining the disks of the second system as a RAID-Z2 pool on a single system and using the second system as source of spare parts or replacement in case of something other than the pool/disks failing. That is assuming the disks are of the same type. If they're not then a 2nd vdev or even two RAID-Z1 zpools with local replication would be better. (RAID-Z2 for preference because see Cyberjock et. al. about how long 2TB+ drives (or bigger drives with at least 2Tb of data on them) take to resilver, and that's even assuming you swap in a replacement right away.*)

To grossly oversimplify for illustrative purposes, rsync is effectively mirroring (file level) at slow speed to a mirror much further away (non-local). It's (usually) done in this context for physical "disaster" survival reasons where a single physical catastrophe damages the entire facility. "Facility" can be a Data Center or your home, the principle is the same. If you're not doing the physical separation for this reason then you're mathematically doing the wrong thing by having two mirrored RAID-0 pools. Even if you are doing the separation for DR, are you aware what you're trading off: How much data is generated in a week that you can lose due to a simple disk failure vs the relative odds of a facility disaster? (A facility disaster is still less likely even if you live/work in a combat zone or behind a levy or whatever.)

It's not rocket science: If you have something with a defined probability of failure and you have two of them, you just doubled the chances of having a failure. It's not quite linear from there, but as a more extreme example: If you have 1000 devices with a 100,000 hour (~11.5 YEARS) Mean Time Between Failure rating, you will have on average approximately one failure a month. Failure rates are not a bell curve either, they tend to be a bathtub.

Now for the non-math part: Murphy (the patron saint of Engineers) says that a failure will occur at the most critical and unplanned-for location/time. For your reference:

Finagle's corollary to Murphy's Law is usually rendered:
“Anything that can go wrong, will—at the worst possible moment.”

O'Toole's Corollary of Finagle's Law favored among hackers is a takeoff on the second law of thermodynamics (also known as entropy):
“The perversity of the Universe tends towards a maximum.”





* sidebar: practice replacing a disk in a pool before you have to do it for realsies, your data will thank you.

 

[Knowltey]

I understand the reasoning for ZFS and raidz for data protection, but I am after maximum space.

With the size of HP Microservers 7 or 8 drives is the maximum. I use 7 drives split between 2 pools and the 8th drive for jails. I am currently using 4x3tb raid-0 10.7tb total 76% used, 3x3tb raid-0 8tb total 80% used. 1x2tb jails. Losing one drive minimum from either pool will leave my current storage needs short.

Moving up to 5 or 6tb drives around xmas for both servers.

Have prob had 2 drive failures over 6 years, and rsync r done weekly.

THat's the thing though. You already are in a sense running a mirror, you're just doing it the long way between two machines. So even now you aren't currently getting maximum space.

 

[cyberjock]

wow.. I'd *never* run ZFS without redundancy. All you need is some corruption in one pool to be replicated to the other thanks to zfs snapshots and you'll be very unhappy when both pools go down at the same time.

 

[SLIMaxPower]

wow so much h8 from everyone.

i loob freenas and how it has performed for me since using from 8* have NOT had any major issues apart from user error.

micro-servers can only fit 6 3.5 drives so that's the limiting factor here + cost/mb issue. I actually have 8 hds, 7 between 2 pools and 1 for jails only. manual rsync weekly to another local micro-server powered on for that purpose only.

if it was only 1 server, then yes of course i would use parity.

cheers

 

[Knowltey]

wow so much h8 from everyone.

i loob freenas and how it has performed for me since using from 8* have NOT had any major issues apart from user error.

micro-servers can only fit 6 3.5 drives so that's the limiting factor here + cost/mb issue. I actually have 8 hds, 7 between 2 pools and 1 for jails only. manual rsync weekly to another local micro-server powered on for that purpose only.

if it was only 1 server, then yes of course i would use parity.

cheers

Yeah, you may not have had a problem so far, but you're just asking for one with everything you're doing regarding the storage configuration. If you're comfortable and aware of the risks then be our guest, but just don't expect remore when you are following the "worst practices guide." We're just letting you know the "best practices guide."

 

[SLIMaxPower]

5 yr and counting :thumbup:

 

[Knowltey]

5 yr and counting :thumbup:

And a higher likelihood of catastrophic failure with every day that goes by :thumbup:

 

[9C1 Newbee]

Do you know who Weird Al Yankovic is? You should cause you need some parity in your life.

 

[Knowltey]

Do you know who Weird Al Yankovic is? You should cause you need some parity in your life.

That or he just needs to at least look in a mirror.

 

[jgreco]

Do you know who Weird Al Yankovic is?

Its all about the Pentiums, baby!

Al's great. I've been to more of his concerts than any other performer... isn't he supposed to start touring again soon?

 

[Knowltey]

Its all about the Pentiums, baby!

Hmm, that does hold true when you're talking about a cost efficient FreeNAS build that supports ECC. It's all about the Pentium (G####) processors.

 

[DaNilePharaoh]

Guys! I salute your optimism, can we change the subject please? I'm about to finish my first build :P