Data: DEGRADED on RN104

As far as the lost connections go, it is best to follow whatever advice mdgm gives you after he analyzes your logs.

njpryan wrote:
I have noticed of late that the space available has suddenly increased from 2GB (approx) to 4GB (approx) for no apparent reason.

Do you really mean GB? or TB?

Do you have snapshots enabled?

njpryan wrote:
With regards to the earlier conversation about RAID-1 and RAID-5 when I add the third 4TB drive - what is going to happen - e.g. will I have 12TB of storage or 8TB of storage and 4TB(slot 2) remains backup of Slot 1 - sorry in advance for seeming confused and repeating some questions.

With xraid2, 2x4TB uses standard RAID 1, and gives you a 4 TB volume. When you add a third 4 TB drive it is automatically converted to RAID-5, and you end up with an 8 TB volume with single redundancy. Slot 2 will no longer be a mirror for slot 1 (as I said above, raid-5 redundancy is more complicated than that). Instead 4 TB's worth of redundant parity blocks are distributed across all 3 drives. This allows any failed (replaced) drive to be reconstructed from the other two.

The basic idea is that if Sector X on drive 1 and drive 2 are allocated for data, then Sector X on drive 3 is allocated for parity (and is the xor of the sectors X on drives 1 and 2). xor is a mathematical operation ("exclusive or") that operates independently for each bit in the sector.

0 xor 0 = 0
0 xor 1 = 1
1 xor 0 = 1
1 xor 1 = 0

It turns out that if

a xor b = c
then
a xor c = b
and
c xor b = a

So you can reconstruct any one of the sector X's from the other two. The parity block pattern is rotated (so 1/3 of the parity blocks are on each drive).

This recovery operation scales to 4 or more drives. Because if
a xor b xor c = d
then
a xor b xor d = c
and
a xor d xor c = b
and
d xor b xor c = a.

So again, any single block that is missing in the "stripe" and be reconstructed from the remaining blocks.

If two or more blocks are missing in the pattern, then RAID-5 can't fix it.

Also, note that when you replace a disk, RAID-5 knows for certain which blocks need to be reconstructed. But if one of the blocks is somehow silently damaged (but can be read) then a xor b no longer equals c. RAID-5 can't figure which one is corrupt w/o help. The bit-rot protection feature in OS6 can (sometimes) figure out which block is bad, and reconstruct it according to the above formulas.

This is also why disk cloning is a last resort when repairing raid - it conceals which blocks are damaged, so there is almost always some file system corruption that RAID cannot repair.