1

I have every 30 minutes smartd messages on /var/log/messages:

smartd[3588]: Device: /dev/sdc, 176 Currently unreadable (pending) sectors

This drive (sdc) is part of RAID 5 configured with mdadm. Mdadm monitor tells RAID is ok but i want to know if I need to change the drive or not. Also if its neccesary to mark as bad this sectors or OS already did it.

If I need to change the drive, how can I chose the replacement one? I can´t find the number of blocks in hard drive specifications so if I chose one with less blocks than original, I will be in trouble.

Rohit Gupta
  • 356
  • 2
  • 4
  • 14
Abraham
  • 73
  • 1
  • 1
  • 13

2 Answers2

2

The accepted answer is generally good, but since mdadm 3.3 you can replace a drive in the array without removing the faulty one first. If the faulty drive mostly readable this protects you from most double failures.

Assuming sdd is a new drive, instead of point 3 and 4 for do this:

  1. Add new drive to array and replace the faulty drive to the new one: mdadm /dev/mdX -add /dev/sdd --replace /dev/sdc --with /dev/sdd

It will trigger a rebuild with with all the possible data avaiable, and will remove the faulty/old drive when ready.

On additional note when creating a new array it's a good practice to create 1 big partition on every physical device, and create the raid array on the partitions, rather than on the raw device.

Daniel Papp
  • 31
  • 3
2

Yes, change the drive.

Unreadable (pending) sectors are sector whose contents could not be read. On a normal non-RAID situation that would result in either a read error, or a long delay while the drive attempts to read the sector again and again until it succeeds (or until it eventually gives up).

With RAID two things are happening:

  1. Your disk is probably configured with a short TLER value. Thus is will give up its attempts to read that sector within a reasonable time. (Thus preventing long hangs).
  2. Your RAID array notices the failure and reads the data from another disk. This is the advantage of RAID 5; you have a spare copy.

What you want to do is:

  1. Check your backups. You should not need them if all goes well.
  2. Fetch a replacement disk of equal or larger size. You can check the size with smartctl -a /dev/sdc. Do not assume all drives of size X have the same capacity. Manufacturers like round numbers; one 500GB drive might well be smaller than another 500 GB drive.
  3. Bring the disk with problems off-line. (mdadm --manage --remove /dev/mdX /dev/sdc)
  4. Replace the disk with new hardware and let the array rebuild itself. (mdadm --add /dev/mdX /dev/sdc)

If you used large disks then this will take a lot of time. Sometimes it is faster to just rebuild the RAID array from scratch and restore from backups. (TEST those backups first!)

While the RAID is rebuilding you have no redundancy. Thus is another disk fails (e.g. due to the stress of rebuilding) then you have a problem. This sometimes happens with large disks (long rebuild times) and batches of drives from the same date.

Hennes
  • 4,842
  • 1
  • 19
  • 29