Why should I not monitor S.M.A.R.T on a hardware raid controller?

[AndyJoiner]

The wiki (http://doc.freenas.org/index.php/S.M.A.R.T._Tests) states:

to prevent problems, do not enable the S.M.A.R.T. service if your disks are controlled by a RAID controller as it is the job of the controller to monitor S.M.A.R.T. and mark drives as Predictive Failure when they trip.

Please can someone elaborate on this statement?

I might be willing to dispense with predictive failure if I can detect an actual failure using SMART and replace the drive (I have a warm-standby freenas, should I loose the entire array).

 

[Nindustries]

Well, the point is: if a RAID controller is controlling your drives, FreeNAS can only see the virtual drives presented by the controller.
So if FreeNAS would be handling SMART, it would never see anything! Because your controller gets the SMART data, but doesn't passes it with your virtual drives.
Conclusion: if you really want to use a RAID controller, let the controller handle the SMART checks. It's the only way anyway.
btw) any sane person would never use a raid controller with freenas. Only HBAs.

(wow, fastest response ever?)

 

[AndyJoiner]

Yup - that was a fast response :)

I'm told that my RAID controller does actually aggregate the SMART data and pass it up the stack. I'll try to verify this when I get my hands on it, hopefully later today.

I have also got an HBA, but there doesn't seem to be a suitable software RAID option that can match the capacity AND redundancy of the existing RAID 5 4x600GB=1.8TB setup (RAID 10 and RAID-Z cannot guarantee more than 1.2TB with those disks).

The overall problem I'm trying to solve (without buying 2 new servers or replacing 8x600GB 15k SAS) at http://serverfault.com/q/585823/34396

 

[AndyJoiner]

p.s. I'm hoping that the RAID controller will present an HDD failure as some kind of SMART error at the array level, e.g. a high Reallocated Sectors Count for the array.

 

[cyberjock]

AndyJoiner, two people have come to me for data recovery in the last week. Both are because they wrongly chose to use a RAID controller. If you are planning to use a RAID controller, just stop that plan right now. It'll end in lost data. Neither of them got their data back, and I don't expect it to change for you either. One guy almost lost his job over it.

As I've said many times in this forum... RAID + ZFS = fail. I've also been quoted many times around here as saying something like "If I were a boss and asked you to build a ZFS server and you included a RAID controller in your parts list to me you would be fired immediately with no questions asked and no chance to defend yourself. That kind of ignorance is just not appropriate where data is important".

 

[AndyJoiner]

Thanks cyberjock. Losing this RAID array will definitely be inconvenient and painful, but it's something we anticipate - we have another layer of redundancy and 2 layers frequent transactional backup. I'm just constrained by budget and free capacity on existing hardware, so I'd really like to use RAID5.

 

[ser_rhaegar]

If you're dead set on raid5, use a different software stack that plays well with a hardware raid controller.

You could also look at raidz which should net near 1.8TB. However, like raid5, it will be a disaster during a rebuild if you hit a URE. Though your drives are 600GB, so presumably they are enterprise level with a URE of 1E15 or higher. You'd have to double check.