2

which two disks can I remove from RAID 10 without breaking the array?

near-copies means that there is raid like this? 

raid0(raid1(sdaX, sdbX), raid1(sdcX, sddX))

= sdb and sdd can be removed

cat /proc/mdstat 

Personalities : [raid1] [raid10]
md2 : active raid10 sda3[0] sdd3[3] sdc3[2] sdb3[1]
      19550976 blocks 64K chunks 2 near-copies [4/4] [UUUU]

md1 : active (auto-read-only) raid10 sdc2[0] sdd2[3] sda2[2] sdb2[1]
      7823488 blocks 64K chunks 2 near-copies [4/4] [UUUU]
Filip Svoboda
  • 23
  • 1
  • 3
  • i was under the impression raid 10 was a raid 1 of raid 0 arrays, so raid 1(raid0()raid0()), which isn't what you've stated ? Isn't this raid 0+1 ? Outcome should be the same, you can remove one from each raid 1 as its mirroring, but not more than one from each, as that'll break the raid 0. – Sirex Nov 22 '10 at 14:56
  • @Sirex In the Linux Kernel, there is a separate profile called raid10 that functions like a raid0+1 but is a single array (rather an array of arrays) and is a bit more flexible than raid0+1. More info: http://neil.brown.name/blog/20040827225440 – Ophidian Nov 22 '10 at 15:24
  • 1
    Take the server offline, unplug two. Turn it on again, if the RAID can't be mounted then you picked wrong, unplug one, and reconnect one of the first two, you should have found your two drives. In any case, I how you have a good backup of your data, whatever your doing sounds dangerous. – Zoredache Nov 22 '10 at 17:37
  • @Sirex Raid 1+0 is a stripe of mirrors. The first number is the first operation applied (mirroring) and the second number is the second operation applied (striping). http://en.wikipedia.org/wiki/Nested_RAID_levels#RAID_10_.28RAID_1.2B0.29 – Mark Wagner Nov 22 '10 at 20:16

2 Answers2

3

Given that you have 4 drives and are using near copies your RAID10s degenerate into RAID1+0, i.e., a stripe of mirrors. See, e.g., this posting which is straight from the horse's (Neil Brown's) mouth.

However, for your RAID10s the drives are ordered differently:

 md2: sda3[0] sdb3[1] sdc3[2] sdd3[3]
 md1: sdc2[0] sdb2[1] sda2[2] sdd2[3]

This means you could lose sdb and sdd and be OK but you could not lose any other combination of 2 drives.

Skyhawk
  • 14,200
  • 4
  • 53
  • 95
Mark Wagner
  • 18,019
  • 2
  • 32
  • 47
  • Thank you. Why I couldn't lose sda and sdc? in md2 remains sdb3[1] from the first mirror and sdd3[3] from the second mirror and in md1 remains sdb2[1] from the first mirror and sdd2[3] from the second. – Filip Svoboda Nov 23 '10 at 18:22
  • Oops, you're right. :) – Mark Wagner Nov 23 '10 at 23:11
  • 1
    Still, that's some scary stuff. If mdadm had done abcd / bcad, you'd be wishing you had raid10 all and then built partitions on top. I wonder if it's a lucky shot that it just switched a/c. – Morg. May 29 '14 at 10:40
1

For a RAID 10 (rarely seen documented as 1+0) array you can lose 1 disk (doesn't matter which) in each RAID 1 array and still be functional. Unfortunately many manufacturers cause confusion between RAID 10 and RAID 0+1.

murisonc
  • 2,968
  • 2
  • 21
  • 31