How to create a software Raid 5?

[Vector]

Wow, it sounds like you guys dug a LOT deeper that i did, and that this problem may go further than just the power supply cable arrangement, etc...

@ Flan: I can't find my receipt for the power supply, so i can't tell who/what it is. I threw all of the boxes for my computer parts away about 2 or 3 months after i built it. The next time i open it up to clean it i will try to remember to take the power supply out and see what it says on the sticker. I have a http://www.google.com/search?q=silve...aven+RV01+case so i can't see the label when i open the door because it's covered up by the cases frame.

[JaXXoN]

Just to keep you in the loop, I repeated the installation (as describe in my
previous posting) and everything seems to work fine, now. I'm still not sure
what wiped out /boot in my previous attempt but I wouldn't exclude a pilot
error :-)

regards

Bernhard

[JaXXoN]

It did it again! Here's what I could capture:

/dev/md127: Superblock has an invalid journal (inode 8)
      CLEARED
*** ext3 journal has been deleted - filesystem is now ext2 only ***

/dev/md127 was not cleanly unmounted, check forced
/dev/md127: Resize inode not valid.
/dev/md127: UNEXPECTED INCONSISTENCY ; RUN fsck MANUALLY.

When then mounting /dev/md127, then the filesystem is pretty much
corrupt (you can see a lot of "??????" when doing an "ls").
Fortunately, I had a backup of /dev/md127 (/boot) so I could
easily restore my system. However, I fear this is going to happen
again, soon.

Seems like something goes wrong during system shutdown.

Here's my /proc/mdstat

Personalities : [raid6] [raid5] [raid4] [raid1] 
md126 : active raid1 sdb1[0] sde1[3] sdd1[2] sdc1[1]
      1048564 blocks super 1.0 [4/4] [UUUU]
      
md127 : active raid5 sdb2[0] sde2[4] sdd2[2] sdc2[1]
      2927135232 blocks super 1.2 level 5, 512k chunk, algorithm 2 [4/4] [UUUU]
      
unused devices: <none>

(BTW.: md126 and md127 are flipping around, means some md126 is the raid1
and sometimes md127 is the raid1)

/etc/mdadm.conf:

ARRAY /dev/md/127 metadata=1.0 UUID=b26ee822:186c373e:f58450a0:a7539a94 name=bee.qnlx.de:127
ARRAY /dev/md/126 metadata=1.2 UUID=eb112468:bfb44aa9:0806fbe9:788ec074 name=bee.qnlx.de:126

feedback is highly appreciated!

regards

Bernhard

[Vector]

I know that this does not help, and it's probably also not what you want to hear, but that's one of the reasons that i don't raid the OS drive itself.

i've got 5 hard drives: 4-2tb drives (8tb total), raided as a single 6tb drive. Then i have a 1.5tb drive that is the OS drive, and is NOT raided at all. I don't care about the OS drive. If something happens to the os, i'll just reinstall. All of my data for all of my servers is written to the raid anyway. But the point is, less complexity on the OS drive. Less aggravation.

Anyway, fwiw...

[JaXXoN]

Originally Posted by Vector

I know that this does not help, and it's probably also not what you want to hear, but that's one of the reasons that i don't raid the OS drive itself..

OS mdraid worked fine me for about 15 years, so if it doesn't do that today, then
that's clearly a regression! Having SW raid not working reliable is not an
option for me :-) In this context, I'm having the impression that udevd is
doing more harm than good in general, but YMMV ...

Anyway, here's an update:

When doing a shutdown with "poweroff", then just before the PC switches
off, I can read a "Stopping md devices" (or something like this). When using
"halt", then there's also the stopping message, but afterwards there are
"syncing disks" messages in addition (one for each drive), then the
system powers off.

So it seems that "poweroff" turns of power too early (interrupting the
syncing process), and "halt" shouldn't power-off the system at all
(just keep it powered after shutting down the kernel).

While "halt" seems to do the right thing for me, now (stopping MDs,
syncing disks, then power-off), i guess this need to be fixed.

regards

Bernhard

---------- Post added at 11:43 AM ---------- Previous post was at 11:24 AM ----------

Originally Posted by Vector

I know that this does not help, and it's probably also not what you want to hear, but that's one of the reasons that i don't raid the OS drive itself.
[...]
the point is, less complexity on the OS drive. Less aggravation.

I'm basically also a strong supported of the KISS principle ("keep it simple,
stupid"), but as already mentioned, mdraid has been around for a while
with an excellent track record (at least in my book), so my intention is
to create an even more complex setup: I'd like to add a SSD to form
a hybrid RAID1 where read operations are most typically carried out
by the SSD. The RAID1 is nested with RAID5 on the HDD side like this:

md0 = sda1 + md1      /         RAID1 (SSD+RAID5)
md1 = sd{b,c,d,e}1              RAID5
md2 = sd{b,c,d,e}2    /home     RAID5
sda            SSD
sd{b,c,d,e}    HDDs

Let's where this is going to take me :-)

Bernhard