Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

Intel Matrix Storage Console 8.9 showed a degraded array with one disk failure. Yet it offers the option to mark the disk as ok and rebuild the array? When would it be appropriate to do this? Does it assess disk failure incorrectly? Why offer this option?

This is a test server, and I have backups, so am not terribly concerned and tried marking the disk as ok, and it rebuilt the volume without indicating a further problem. BUT is there a problem anyway?

Additionally...

The great responses make me wonder, what the best methods to test the disk might be. SMART tests are mentioned below. Probably I will remove the drive, rebuild with a new one.

It still seems unclear to me whether a volume can rebuild and not show errors, as appears to have happened already with this existing drive?

share|improve this question
1  
If you have backup I think you should delete the array and start again. –  Ricardo Polo Oct 17 '11 at 23:26
1  
That's a little extreme. Why not just slot in a new drive and let the array rebuild on it? –  sh-beta Oct 17 '11 at 23:43
1  
Sure it's safe, just like walking across a busy freeway while wearing a blindfold. Don't mark a disk as OK unless you KNOW it's OK, –  John Gardeniers Oct 18 '11 at 3:57

4 Answers 4

up vote 10 down vote accepted

Drives can be marked as failed in an array for many reasons. Maybe there's a few defective sectors. Maybe the drive heads are failing. Maybe cosmic rays hit your drive at the right angle and time to fail a scan. Maybe their firmware has a bug that breaks under .

Some of these are reparable failures, some aren't.

The thing is, it's really hard to predict hard drive failures. Google's infamous paper found that SMART was only useful in that if it alerted, the drives were more likely to fail than if it didn't. Fully 36% of the failed drives had no SMART errors, fatal or not. So you could run a full suite of SMART scans, find none, and know no more than you do now.

But, assuming this was an out-of-the-blue failure and not an I-did-something-funny-and-it-failed failure, you already have an indication of problems with the disk. Now it's a question of value.

  • How much does another drive cost?
  • How much time would be lost for its users if this server died?
  • How much of your time would be lost if this server died?
  • How much is all that time worth?
  • Double this value to account (naively) for opportunity cost

I've never been in a situation where it was worth letting a drive fail. Why go through the pain? Chances are, the drive you need is pretty cheap. Just buy it and move on.

share|improve this answer

I once had a faulty caddy in an old U160 SCSI array, that was one of 14 disks in the array. When I replaced the caddy (the disk was fine), it still thought it was failed because the disk had the same serial number.

So I marked it as OK, the array re-built and all was fine until we de-comissioned it.

It all depends on your situation, but normally I would never mark a disk as OK unless I was 100% certain that it was OK. Even at 99.9% certain, I would delete the array and start again.

share|improve this answer

If you care about the data, replace the drive immediately with a new one and rebuild the array. You can then run extensive testing on the removed drive and requalify it for use if it passes. However, if you try to rebuild the failed drive in place, you are extending the time you are vulnerable to a double-drive failure should something go wrong during or after the rebuild process.

share|improve this answer

It entirely depends on the reason the drive was failed. In some cases ive seen perfectly fine disks get failed on startup with cheap raid cards because the controller had a derp moment and didnt detect the drive. This is pretty rare though, i ran a bunch of SMART tests on the drive and did a full badblocks test run through by wiping the entire drive with DD. That particular drive was ok by all my standards and as i was running raid5 and not Linear or raid0 i added it to the array again.

Run a SMART test using a Linux recovery disk or similar, make note of the badblocks count, run a full SMART test and then look at the bad blocks count again. If it spiked by anything more then 20 i wouldn't trust it. Same if the badblocks are particularly high for that drive size/make.

The risk is not just that the drives completely fail but that your data may corrupt over time.

Can you also include the readout of "smartctl -a /dev/hda" for this drive in the original question thanks.

share|improve this answer
    
Only reason i would even entertain the idea of using the previously failed drive is because its a test server and you dont appear to care for its data. If it was a production server i would bin the drive and replace it, not questions asked. –  Silverfire Oct 17 '11 at 23:53

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.