mentby.com
Blog | Jobs | Help | Signup | Login

Why does GENERIC kernel for OpenBSD 4.8 and 4.9 not support software RAID



I've been using the very helpful notes at http://www.eclectica.ca/howto/openbsd-software-raid-howto.php,  which
works, but does document needing to build a new kernel to enable
software RAID. (I included GENERIC.MP in my GENERIC.MPRAID
configuration: it's a multi-core machine.) But I had a fascinating
chat with another engineer, who I suspect has worked more with OpenBSD
than I have, who is very surprised at this requirement. He pointed out
that GENERIC is normally the "kitchen sink" kernel, with everything
possible enabled.

I've checked the source, and I see where the RAID is disabled in the
GENERIC kernel, but would like to know why. Is there a significant
performance or reliability risk with that feature enabled? I've not
found anything in the archives that explains this, although it's
*awfully* hard to find useful information with such generic keywords
as "software RAID".....


Nico Kadel-Garcia Wed, 04 May 2011 05:56:20 -0700

* Nico Kadel-Garcia <nkadel*******> [2011-05-04 14:55]:

they are not helpful, they are 100% obsolete.

you want softraid(4), which is, surprise! in GENERIC.

--
Henning Brauer, hb*******
BS Web Services,  http://bsws.de
Full-Service ISP - Secure Hosting, Mail and DNS Services
Dedicated Servers, Rootservers, Application Hosting


Henning Brauer Wed, 04 May 2011 06:05:59 -0700

The one of the many benefits of OpenBSD is its superb documentation
especially when comparing with lack of that in Linux and other systems
which tries to supply that by posts around Internet which are outdated
and mostly wrong.

So see  http://www.openbsd.org/faq/faq14.html#RAID


Tomáš Bodžár Wed, 04 May 2011 08:02:38 -0700

I still use raid(4) -- RAIDframe -- for it's root-on-RAID capability.  I
eagerly await the completion of root-on-RAID with softraid(4).  

My thanks to Joel, Jordan, Marco, and the rest of the team developing this.


Josh Grosse Wed, 04 May 2011 10:09:37 -0700

Not "everything possible"; rather "everything we expect to
work reasonably well".


Stuart Henderson Wed, 04 May 2011 15:25:29 -0700

I use RAIDFrame too, but it was a mistake; I had no idea RAIDFrame was
no longer maintained and had no idea of the existence of softraid when I
installed and implemented 4.6/4.7 machines over the last year or so.

I hadn't used OpenBSD since 3.x days but I knew I needed it for some
routing at work. I also knew I needed some software RAID. Almost every
result from Googling something along the lines of "installing OpenBSD
onto a software RAID" leads to a RAIDFrame guide like the one at http://www.eclectica.ca/howto/openbsd-software-raid-howto.php

I'm excited to hear softraid is coming along, and remember reading that
support for booting off of one was recently committed. The work done on
softraid is very appreciated and I look forward to seeing more of it
committed, but the reality is there is a significant split regarding
software RAID in OpenBSD. I bet I'm not the only person using RAIDFrame
close to production without realizing it's not even maintained code.

I think this is mainly due to the fact that softraid can't be used for
the root partition (or booted off of, for now). This leads everyone to
follow RAIDFrame guides to install OpenBSD onto software RAID1, but
nobody bothers to mention that RAIDFrame isn't actually maintained anymore.

And I have a feeling it's why my routers crash once every few months or
so with some odd, sd0/sd1 related FIFO errors (using SSDs too...). I'm
currently pulling RAIDFrame out of various routers and not using any
RAID at all anymore -- CARP + pfsync + duplicate hardware is enough for
what these routers do.

In no way am I blaming anyone here -- it's obviously my fault that I
didn't read the 4.7 FAQ closer and learn about softraid -- but I think
large amounts of people are being lead to RAIDFrame via Google without
fully realizing what they are using or why they might be making a bad
decision.


Tyler Morgan Wed, 04 May 2011 15:44:14 -0700

(plz all stop pushing that links search engine rank.)

Along the line of: "Why tf doesn't my root-partition change often
enough, so that i am not ok by simply using ALTROOT? Didn't expect me
to read the afterboot manpage as i was prompted, didn't you?"

(Not addressing you personally, just picking up the vibe.)


Robert Wed, 04 May 2011 16:06:47 -0700

Top posting only because this phone's current Email client forces me to.

I think that admonition looks like it was in regards to my post. A ">" or two
are absent. The link was Tyler's, not mine.

--
Sent from my Android phone with K-9 Mail. Please excuse my brevity.

pushing that links search engine rank.) > I think this is mainly due to the
fact that softraid can't be used > for the root partition (or booted off of,
for now). This leads > everyone to follow RAIDFrame guides to install OpenBSD
onto software > RAID1, but nobody bothers to mention that RAIDFrame isn't
actually > maintained anymore. > > And I have a feeling it's why my routers
crash once every few months > or so with some odd, sd0/sd1 related FIFO errors
(using SSDs too...). > I'm currently pulling RAIDFrame out of various routers
and not using > any RAID at all anymore -- CARP + pfsync + duplicate hardware
is > enough for what these routers do. > > In no way am I blaming anyone here
-- it's obviously my fault that I > didn't read the 4.7 FAQ closer and learn
about softraid -- but I > think large amounts of people are being lead to
RAIDFrame via Google > without fully realizing what they are using or why they
might be > making a bad decision. Along the line of: "Why tf doesn't my
root-partition change often enough, so that i am not ok by simply using
ALTROOT? Didn't expect me to read the afterboot manpage as i was prompted,
didn't you?" (Not addressing you personally, just picking up the vibe.)


Josh Grosse Wed, 04 May 2011 20:37:18 -0700

* Tyler Morgan <tylerm*******> [2011-05-05 00:43]:

if "it's not in GENERIC" is not a strong enough hint, I dunno.

--
Henning Brauer, hb*******
BS Web Services,  http://bsws.de
Full-Service ISP - Secure Hosting, Mail and DNS Services
Dedicated Servers, Rootservers, Application Hosting


Henning Brauer Wed, 04 May 2011 23:31:15 -0700

BTW, I setup a server with ALTROOT feature. After 3 months, the root
disk (where / lives) partially broke and the daily job screw up my
/dev/r$rootbak with the dd(1) command ...

--
Olivier Cherrier - Symacx.com
mailto:oc*******


Olivier Cherrier Thu, 05 May 2011 01:02:09 -0700

Welcome to the world of RAID.

I suspect your intent on posting was "hey, altroot ain't so cool, look,
it can fail!", but your REAL lesson should be, "be ready with a backup,
RAID won't always save you".  If you haven't seen similar failure mode
in HW or SW "true" RAID, you have a few choices:  1: consider yourself
lucky, 2: wait for it, 3: die soon (I suggest hoping for option 1, plan
for option 2).

One example: drive failed on machine.  Replace drive.  Remirror (in HW
RAID) fails with message of something like "failure rebuilding new
disk".  Replace new disk, try again, same error.  Repeat a few more
times, until finally realizing the REAL problem was the remaining disk
had an unreadable section on it, and THAT caused the remirror failure.
At least you quickly understood what happened, we lost a couple evenings
working on that one.

Nick.


Nick Holland Thu, 05 May 2011 04:14:38 -0700

Give the man a kewpie doll!!!

I'm dealing with modest server hardware, which didn't have workable
hardware RAID. I've found the hardware RAID compatibility chart to be
awkward: much of the recommended hardware is no longer manufactured or
not sufficiently specified to rely on. Naming a chipset is not enough:
a model number is really ideal, because saying "LSI" or "Dell Perc"
covers dozens of different cards, some of which may be very reliable
but others are not. I used to recommend 3Ware from good experience and
and general UNIX/Linux compatibillity, but they got bought by LSI:
gods only know what their quality is these days.

Also, I just went and tried the sensible guidelines at http://jpiasetz.tumblr.com/post/483365684/software-raid-on-o[..]
which use softraid for non-/boot partitions. Seems reasonable, but
this command simply fails:

    # bioctl -C force -c 1 -l /dev/wd0d,/dev/wd1d softraid0
    # dd if=/dev/zero of=/dev/rsd0c bs=1m count=1
    bioctl: Can't locate raid0 device via /dev/bio

The documentation on bioctl is reasonably, and detailed, and not very
useful due to not having examples for noobs.


Nico Kadel-Garcia Thu, 05 May 2011 05:00:58 -0700

kind of the point.
the examples explain the syntax in a more readable way.
the manpages encourage to understand how the magic works,
instead of just blindly pressing the blue button.


Robert Thu, 05 May 2011 09:19:15 -0700

Actually, it's a little known fact that all the best bits of code are
automatically turned on by an email from theo when you've bought your
50th cd and when you've used 50 downloaded cd's without buying one, an
email changes your language to klingon.

Top Tip: you can cheat the system by buying 50 4.9s, you don't have to
wait for 50 releases.


Kevin Chadwick Thu, 05 May 2011 12:24:08 -0700



Related Topics

Post a Comment