gmirror losing drive

[Paul Mather]

On Tue, 19 Apr 2005 14:38:33 +0200, Andrea Venturoli
<ml.diespammer at netfence.it> wrote:

> Danny Howard wrote:
> 
> > I'm not entirely sure on this one ... you have RTFM?
> 
> Obviously. I started with the tutorial at 
> http://people.freebsd.org/~rse/mirror/, but I read the whole manual 
> before setting it up.
> 
> 
> 
> > There's "gmirror 
> > configure -a" ... but that is about synchronization.
> 
> Yes, and synchronization works fine.
> 
> 
> 
> > What does gmirror info says before you forget / insert?
> 
> Hm, I'd have to reboot.
> 
> Right now it says:
> 
> Geom name: gm0
> State: COMPLETE
> Components: 2
> Balance: split
> Slice: 2048
> Flags: NONE
> SyncID: 5
> ID: 2253479574
> Providers:
> 1. Name: mirror/gm0
>     Mediasize: 36778544640 (34G)
>     Sectorsize: 512
>     Mode: r3w3e2
> Consumers:
> 1. Name: da1
>     Mediasize: 36778545152 (34G)
>     Sectorsize: 512
>     Mode: r3w3e3
>     State: ACTIVE
>     Priority: 0
>     Flags: DIRTY
>     SyncID: 5
>     ID: 4069582681
> 2. Name: da0
>     Mediasize: 36778545152 (34G)
>     Sectorsize: 512
>     Mode: r3w3e3
>     State: ACTIVE
>     Priority: 0
>     Flags: DIRTY
>     SyncID: 5
>     ID: 4172309470
> 
> Geom name: gm0.sync
> 
> 
> After a reboot I only see da1 under Consumers and State is DEGRADED 
> (Componentis are still 2, though).
> 
> 
> 
> > My hunch is that you are not rebooting cleanly, so when the system
> comes 
> > up, gmirror thinks it has to re-sync the disks
> 
> I would expect this behaviour, but:
> a) I am rebooting cleanly;
> b) it doesn't just need resync (that happened to me on another
> machine), 
> it really loses one drive/Consumer!!!

> BTW: system is 5.3p9 now

I've been using geom_mirror for quite a while.  I currently use it on
6-CURRENT and 5.4-STABLE systems.  The implementation has evolved since
first I used it, and some kinks have been worked out.  You may be
experiencing some of these kinks.

Do you have swap on your geom_mirror?  In earlier versions of
geom_mirror, even a clean shutdown would cause a mirror to be marked
degraded if swap was still active.  The fix at the time was to amend the
stop_cmd in /etc/rc.d/swap1 to do an explicit "swapoff -a" (or similar).
Later modifications rendered this explicit step unnecessary.

I also remember a short window in which I experienced the same problem
of losing a drive/consumer at reboot when a mirror would become degraded
(in my case, due to the TIMEOUT - WRITE_DMA problems I was having with
5.x/6.x).  This seemed to go away with a subsequent upgrade.

So, if you are planning to upgrade when 5.4-RELEASE rolls around, you
might find that these problems disappear.  Otherwise, you may be stuck
with at least one of them.

Cheers,

Paul.
-- 
e-mail: paul at gromit.dlib.vt.edu

"Without music to decorate it, time is just a bunch of boring production
 deadlines or dates by which bills must be paid."
        --- Frank Vincent Zappa