On mine I would partition the 8tb into 4tb partitions.
Then replace one of the 4tb partitions with an 4tb partition on an 8tb.
Then create a new 2 disk raid5 array with the old 4tb and the #2 8tb partition
Next add looks similar and you then have 3 disk raid5.
I have my array build of partitions of several disks such that when I
get a bad block on one of the partitions typically only one partition
gets thrown out of its array and when doing ops on the smaller
partitions I don't ever have any operation that takes anywhere close
to a full day.
I am using gpt so I also number the partitions such that they match
the last number of the md device like this:
md13 : active raid6 sde3[12] sdd3[1] sdb3[7] sdh3[8] sdg3[10] sdi3[11] sdf3[9]
3612623360 blocks super 1.2 level 6, 512k chunk, algorithm 2
[7/7] [UUUUUUU]
bitmap: 0/6 pages [0KB], 65536KB chunk
that way I always know 13 gets a *3 device.
The above array was all 3tb disks originally, now 3 of them are
partitions on 6tb disks and I have a 4 disk raid6 added into a
separate fs using the one "good" 3tb disk I removed, and the 3 2nd
partitions on the 6tb disks (the other 2 3tb were having sector read
issues).
This lets me get a bit of extra space without doing an immediate buy
of all 7 new 6TB disks at once.
On Thu, Oct 28, 2021 at 11:44 AM Wol <antlists@xxxxxxxxxxxxxxx> wrote:
>
> On 28/10/2021 14:36, David T-G wrote:
> > Hi, all --
> >
> > It's time to replace a few 4T disks in our little server. I don't
> > particularly want to go back with more 4T disks (although the same
> > model sure are cheap these days! :-) and figure I should put in larger
> > drives as I go. I am then left with weighing simple $/G vs total price;
> > bigger drives can be cheaper per volume but of course more overall.
> > My first approach is to put newer, larger drives in place and expect to
> > grow into the empty space when all of the old ones have been swapped out.
> >
> > But ... If I were to splurge and buy 3ea big drives to replace all of
>
> Does that mean you have three drives currently?
>
> > the space that I have now, how practical is it to grow that RAID5 array
> > by adding additional drives later?
>
> Very. mdadm --add ...
>
> > My eventual goal would be to get
> > to 8-10 devices in a RAID6 layout (two "extras"), but of course I can't
> > afford that today. Do I have an easy path to get there in the long run?
> >
> > [BTW, can I convert an array from RAID5 to RAID6, too?]
> >
> > On the other hand, I do have the empty slots (currently filled with
> > scratch drives here and there), so I could both replace my aging drives
> > and add more and just grow this array 1) if the growth idea is practical
> > and 2) if I don't get to splurge.
> >
> Okay, buy your new 8TB drives in pairs (unless you've got a bunch of
> scratch 4TB drives).
>
> Assuming you've got a 3x4TB array this will get you to a 3x8TB in one hit...
>
> mdadm --replace 4TB with 8TB
> Twice.
>
> mdadm --create --level=striped 4TB 4TB (to give you an 8TB raid0)
>
> mdadm --replace 4TB with 8TB raid0 mirror
>
> You may then be able move the data off your scratch drives onto the
> array, create another 8TB raid0, and add that for a raid6. You can then
> just add 8TB drives bit by bit.
>
> Read the website - there#s a lot of info there...
> https://raid.wiki.kernel.org/index.php/Linux_Raid
>
> Cheers,
> Wol
