1 |
Am Mittwoch, 4. Januar 2012, 22:45:45 schrieb Jeff Cranmer: |
2 |
> On Thu, 2012-01-05 at 04:01 +0100, Volker Armin Hemmann wrote: |
3 |
> > the short one: |
4 |
> > |
5 |
> > partition one disk with (c)fdisk. Use sfdisk to transfer the partition |
6 |
> > scheme to the other disks. |
7 |
> > |
8 |
> > run mdadm --create /dev/md0 level=whatever you want --raid- |
9 |
> > devices=thenumberofdevices /dev/sdXY /dev/sdZY ... |
10 |
> > |
11 |
> > mdadm --detail --scan >> /etc/mdadm.conf |
12 |
> > |
13 |
> > done |
14 |
> |
15 |
> OK, but there is active data on the disks, so I don't want to partition |
16 |
> them. They should already partitioned, and running fdisk will erase the |
17 |
> data. |
18 |
|
19 |
first rule: |
20 |
|
21 |
always mount a scratch monkey |
22 |
|
23 |
In your case: always backup data. |
24 |
|
25 |
There is a way to preserve the data on one disk, create a raid5 with one disk |
26 |
missing, then copying the data onto the raid and add the disk. |
27 |
|
28 |
But that is high risk stuff. |
29 |
|
30 |
> |
31 |
> If I run mdadm --create /dev/md0 level=5 |
32 |
> --raid-devices=3 /dev/sdb /dev/sdc /dev/sdd, will that erase data |
33 |
> already on the disks? |
34 |
> |
35 |
> Prior to running this command, there is no /dev/md entry. Is this |
36 |
> correct? |
37 |
|
38 |
yes. You might have to create the nodes with mknod - my memory is sketchy |
39 |
there. |
40 |
|
41 |
|
42 |
> Looking further by using fdisk, it appears that sdc has a linux |
43 |
> partition on sdc1 starting at sector 34, and a GPT partition of size 0+ |
44 |
> at /dev/sdc4, sector 0. Nothing else is on that disk (no sdc2 or sdc3). |
45 |
> |
46 |
> sdd and sdb report invalid partition table flags and do not appear to |
47 |
> have active partitions. Does this make sense? |
48 |
|
49 |
if you used fakeraid before, yes. But that means: without the original |
50 |
fakeraid everything on that disks is inaccessible... and you need to partition |
51 |
them. |
52 |
|
53 |
> |
54 |
> Is it possible that I ordered the disks incorrectly when I installed |
55 |
> them, and by simply swapping disks b and c at the raid I can get things |
56 |
> to start making sense? Is there an order to a set of RAID5 disks? I |
57 |
> thought any two of three RAID5 disks could be recovered, regardless of |
58 |
> which one dies? |
59 |
|
60 |
no. |
61 |
First, the order of the disks is irrelevant, but the most important thing: |
62 |
|
63 |
with Raid5 ONE disk out of an array might fail. No matter how many disks - two |
64 |
fail and everything is lost. |
65 |
|
66 |
> |
67 |
> > there is a reason why I never ever touch genkernel. |
68 |
> > |
69 |
> > you should forget that crap. You don't need to copy around anything. If |
70 |
> > your root is not on some fancy setup, you don't need initramfs. |
71 |
> > |
72 |
> > Just make a nice kernel, put it in /boot. Done. |
73 |
> |
74 |
> OK. The OS disk is non-RAID (120GB SSD), so I don't need any fancy |
75 |
> options in my kernel. All the domdadm and dodmraid stuff is needed just |
76 |
> when your OS disk is raided. Correct? |
77 |
|
78 |
yes |
79 |
|
80 |
|
81 |
-- |
82 |
#163933 |