1

Lets say I remove a drive from a RAID array on a HA server for a period of time then re-add it. Is there a way to limit the rebuild to a diff of what has changed since it was pulled, or does it have to do a complete rebuild?

We have been using Spinrite on our non-raid devices to catch drive damage before it gets bad enough for SMART to notice them. And it has been great on SSD's running on level 1 or 2, proving bad cells to the internal controller that it was unaware of. What we would like to do is pull a drive from a RAID, run Spinrite on the drive at level 1 (since it cannot isolate it from a hardware RAID), then return it to the RAID. With an SSD rewriting the entire contents of the drive regularly would obviate any reliability gains from proving readability of the drive. When the drive returns is there a way to rebuild it with just the diff of the changes since the drive went out of service instead of the entire contents?

Tvanover
  • 257
  • 1
  • 3
  • 13

1 Answers1

2

If the drive (or the rest of the array) is not carrying metadata which would allow for only synchronizing the changes since it has been pulled, then no, you would not be able to avoid doing a full resync of the drive. So far, the only system I have seen this work is ZFS.

Note that using Spinrite (or any other tool for that matter) to do a full read-write cycle of the data on a pulled RAID array drive is a bad (so not to say braindead) idea. By pulling a drive you are degrading the array, decreasing its performance and unnecessarily increasing the risk of a total array failure on a regular basis. You also should abandon the idea that Spinrite would sprinkle Fairy Dust on your drives so they would live longer and be more reliable.

Most RAID controllers support the "patrol read" or "consistency check" functionality which would read all data on each of the physical disks and verify it against a checksum (if present) without degrading the array - this is what you should be looking at instead. Obviously, the results of those checks would need to be monitored and suspicious drives replaced.

the-wabbit
  • 40,737
  • 13
  • 111
  • 174