Hi everyone!
While setting up a new box w/ RAID I decided to stress-test the latter
BEFORE losing anything important.
Meanwhile I've setup a test VM and did the following.
Install OpenBSD 7.1, nothing special.
I.e. de keyboard, hostname foo, default network cfg, no SSH, no X, no
com0, no users, default timezone, default disk setup, default file sets.
Wipe[1] the additional three disks (the system is not installed on).
Run syspatch.
Run syspatch.
Run syspatch.
Run sysupgrade. -> 7.2
Run syspatch.
Run syspatch.
Reboot[1].
Run disklabel -E for each RAID disk:
D ENTER
a ENTER
ENTER
ENTER
ENTER
RAID ENTER
q ENTER
ENTER
Assemble RAID5: bioctl -c 5 -l sd1a,sd2a,sd3a softraid0
Run disklabel -E sd4:
D ENTER
a ENTER
ENTER
ENTER
ENTER
ENTER
q ENTER
ENTER
Make FS: newfs sd4a
Mount it: mount /dev/sd4a /mnt
Put some files on it.
---
[1] This way the OS definitely sees no MBRs and any new disklabels will
be plain BSD ones. By the way I previously tried GPT (fdisk -gy) and BSD
disklabel on top (disklabel -E) (as I'm on a x86 box), both on the three
disks and on the virtual RAID5 "disk" before creating FS. Writes with F3
or the equivalent,
while dd bs=32M count=32 if=/dev/urandom of=/mnt/`date +%s`; do true; done
, I used to "Put some files on it" freezed the whole pre-7.2 snapshot OS
(even no ddb> ). Sure, why to use (and even support) nested part. tables
on non-boot disks (and I've "solved" the problem by using plain BSD
disklabels). But a complete OS freeze, despite su nobody and renice -n
19? Really?
Btw. thanks to the ones of my BCCed colleagues who let me test this on
several boxes at work. I couldn't reproduce the freeze, but hey I tried
it. (Have I already cursed the one box which hw RAID ctrler doesn't
support OpenBSD?)
---
Reboot.
Mount.
Data is still there.
Good.
Boot a Linux, wipe 1/3 disks.
Boot OpenBSD again.
bioctl sd4 shows degraded; 1/3 disks missing.
Okay...
New disklabel on wiped disk, bioctl -R /dev/sd3a sd4 and get a coffee...
---
[1] In some case, I think 7.1-stable and/or pre-7.2 snapshot with nested
part. tables as mentioned above, sd4 wasn't even present, so couldn't do
bioctl -R. I had to use bioctl -c ... .
---
Rebuild done? Reboot!
Mount the FS again.
Type cat /mnt/<TAB>... er, what?
At this point I actually planned to write about the data still being
there IF(!) you didn't mount anything before[2] rebuild+reboot.
And then I saw this:
https://files.al2klimov.de/s/z8JW8py2KM5nbEB
(Come. On. This isn't even funny. :'( Funny is only that the box I do
all this for shall be the backup server for the box files.al2klimov.de.
So if the latter is gone, the pictures I linked above are also gone
because there's no backup on the server I can't set up until I'm ensure
that RAID is OK which the crash pictures are for.)
Are the screenshots enough? Do you need anything more? The VM is still
on ddb> .
---
[2] Otherwise you should get a 1GB file -not a dir!- on /mnt. But as
you've already read I was interrupted by a crash.
---
Have I already mentioned a pulled out SATA cable (online!) get handled
well? (Degraded, data still there, at least until reboot.)
Best,
A/K
