On 22/06/2025 19:03, Omari Stephens wrote:
On 6/22/25 09:30, Wol wrote:
On 22/06/2025 02:39, Omari Stephens wrote:
|At that point, /proc/mdstat showed the following, which makes no
sense:|
Why doesn't it make any sense? to)
Don't forget a raid-10 in linux is NOT a two raid-0s in a raid-1, it's
its own thing entirely.
Understood. I've been using Linux MD raid10 for over a decade. I've
read through this (and other references) in depth:
https://en.wikipedia.org/wiki/Non-standard_RAID_levels#Linux_MD_RAID_10
My question is this: Suppose you create a 4-drive array. 2 drives are
missing. What data is there to synchronize? What should get copied
where, or what should get recomputed and written where?
Ah - I get that. But presumably raid *always* fires up a resync (unless
told not to) when an array is created - and see below ...
To my understanding, in that situation, each block in the array only
appears in one place on the physical media, and there is no redundancy
or parity for any block that could be out of sync.
No. It's trying to resync a *broken* array. Unless you had explicitly
asked for just one mirror (which I don't think you did). A raid-10 needs
"no of mirrors plus one". I can see why you would think "why doesn't it
just fire up as a plain raid-1?", but I guess it just doesn't. I've
never tried to create an array in a broken state.
When you read from the array, yes, you're going to get interleaved bits
of whatever happened to be on the physical media to start with, but
that's basically the same as reading directly from any new physical
media -- it's not initialized until it's initialized, and until it is,
you don't know what you're going to read.
Yup. Read-before-write is going to return random junk - not particularly
a good idea - but I don't think you can assume that --assume-clean will
actually create a clean array. My fu isn't up to guessing what will
happen when, in particular for higher raids even with zeroed data the
parity will be one. (And I would never assume a new drive was zeroed. To
me it has as much evidence behind it as an old wive's tale. Other people
may have real evidence -I don't.)
What you did worked, and in hindsight you can see how, but with just two
drives, I personally wouldn't even have tried. Anyways, all's well that
ends well, as they say ...
Cheers,
Wol
