problems with GEOM_MIRROR - the solution!

[Krzysztof Ciepłucha]

Pawel Jakub Dawidek wrote:

> It shouldn't if you have more than one disk.
> So let it be clear: there is no synchronization if there is more than one
> disk and timeout is set to 0?
Yes, on 6-CURRENT it seems to work ok.
But I'm not sure about 5.3BETA2 and BETA3 - I have to check it out on 
the second server. It seems that 5.3BETA builds ignore geom-related 
sysctls set in /boot/loader.conf

> +> 1. what was the purpose of the kern.geom.mirror.timeout (set by default 
> +> to 8)? i can't hardly imagine situation where we must wait for some 
> +> disks to start mirror.
> 
> Believe me, there can be many, e.g. mirror on one ata disk (which is
> detected immediately) and one SCSI disk (which is detected after
> SCSI_DELAY delay).

I don't think so, because waiting for SCSI settlement pauses boot 
process and after that geom mirror looks for providers (in this case: 
scsi disks). BTW: 8 seconds geom mirror timeout is less than default 15 
seconds SCSI settlement timeout, so it wouldn't work in that way anyway.

> 
> +> 2. could you add a comand to geom mirror (maybe geom mirror status ?) 
> +> which shows the current state of the mirror (OK, DEGRADED, 
> +> SYNCHRONIZING, FAILED, etc)? now it is a little bit tricky to determine 
> +> the state of a mirror when one disk is missing.
> 
> Yes, I can. There is such field in graid3 already.
> 

Great to hear that.

Did you make any performance comparisions with hardware mirror 
controllers? It shouldn't be any slower unless there is high CPU 
utilization, right?

Do you think it is better to use geom mirror (which is often easier to 
control / manage) than sticking with hardware-based mirror / raid 
controllers (most of them lacks any management software for freebsd) ?

-- 
kris-at-home-dot-pl