1

Given a hot-swappable 15 spindle RAID 10 array with two failed disks. There is one global hot spare, and it is actively "sparing" the first of the two failed disks. The array is running, but has no further redundancy mechanisms.

Onsite tech reports the "spared" disk has a green LED, despite being marked failed in array management software. The "unspared" disk is amber.

In what sequence would you replace the failed disks? Spared first? Unspared first? Both at once?

AndyN
  • 1,769
  • 12
  • 14

2 Answers2

4

Unspared for sure. That mirror is at risk, where the mirror that is rebuilt with the hot spare is redundant.

MDMarra
  • 100,734
  • 32
  • 197
  • 329
  • +1. The hot spare has effectively become the active member, replacing the failed drive. You could leave it like that forever if you wanted to. The "unspared" drive needs to be replaced first. – joeqwerty Jul 21 '10 at 19:39
  • 1
    +1, Replace the obviously dead drive first, one at a time if there's more than one; then fart around with the not-obviously-dead stuff. – Chris S Jul 21 '10 at 20:23
-1

The answer is none. There's no such thing as a 15 spindle RAID10 array. RAID10 requires an even number of disks (spindles).

joeqwerty
  • 109,901
  • 6
  • 81
  • 172
  • I'm guessing one of the drives is the spare, making it an odd number? Also, if you are saying the question is invalid, then the answer should be "unknown" rather than "none" ;) – AaronLS Jul 21 '10 at 19:55
  • Good catch; I spoke imprecisely. There are 15 spindles. 1-14 are assigned to the RAID10 array, and 15 is configured as a global hotspare. Hardware in this case is a Dell MD3000. – AndyN Jul 21 '10 at 19:59
  • I was being sarcastic. Also, the spare drive is a spare drive, it's not a member of the array. RAID10 requires an even number of drives. IMO the answer from MarkM is the simplest and says it all. No offense intended if my comment seems brusk. – joeqwerty Jul 21 '10 at 20:00
  • @AndyN: No offense, I was being sarcastic in my answer. MarkM is on the money with his answer. You definitely want to replace the "unspared" drive first. – joeqwerty Jul 21 '10 at 20:03
  • The way I read the question I assumed it was a 14 data spindles + 1 hot spare (15 spindles) so it could be valid. It is possible to build a RAID 10/1+0 type pattern that works with odd numbers of spindles - IBM's RAID 1E is an example. – Helvick Jul 21 '10 at 20:05
  • @joeqwerty: None taken! My intuition was telling me to deal with the spared one first.. evidently my intuition is a simple FIFO buffer. Thanks all. – AndyN Jul 21 '10 at 20:12
  • @Helvick: RAID 1E is like a staggered RAID 0 with an extra drive and a copy of the data that's being striped also being striped in a staggered fashion across the drives in the array. (At least that's how I understand it). – joeqwerty Jul 21 '10 at 20:15
  • @joeqwerty - that gels with my understanding too. I haven't ever actually used it but it seems that it should behave\perform like RAID 10. "Seems" is a terrible thing to say about something as important as this it has to be said. Mind you I might be missing something :) – Helvick Jul 21 '10 at 20:22
  • @joeqwerty, when creating a Linux software RAID10 you can have an uneven number of drives. Of course the Linux RAID10 doesn't really follow the standard definition. – Zoredache Jul 21 '10 at 21:57
  • @Zoredache: That's weird, but thanks for the info. – joeqwerty Jul 21 '10 at 22:04
  • @Zoredache - It's my understanding that doing that will create the array in a degraded state though, correct? – MDMarra Jul 21 '10 at 22:40