Tom Walsh wrote:

Luca Berra wrote:

quoted

On Sat, Nov 29, 2008 at 04:23:17PM -0500, Tom Walsh wrote:

quoted

<sigh> I need help!

i'll try :(

quoted

I've overhauled the software, replaced the operating system
innumerable times with several Mandriva distros: 2008, 2008.1, 2009.
Removed the Mandriva kernel and compiled a stock 2.6.27.7 from
ftp.kernel.org.  Ran the Seagate SeaTools on all four drives, no
errors.  Ran the Western Digital Date Lifeguard on the two drives, no
errors.  Changed from raid5 to raid10, still resyncs on boot.

https://qa.mandriva.com/show_bug.cgi?id=40023

read the whole of it.

Ahhh, yes!  Thank you!

I found myself heading down that path after successfully rebooting the 
machine 10 out of 10 times using a 2.6.22.9 kernel in the Mandriva 
2008.0 distro.  I then built a 2.6.22.19 kernel in the 2009.0 that would 
resync on every boot.  The problem pwrsisted with the new 2.6.22.19 kernel.

I was reverting back to the 2008.0 distro and building a 2.6.27.7 kernel 
when I saw your reply.  Reading the bugzilla makes a whole lot of sense 
of what I was seeing in the dmesg (re: md not finishing before the 
raid10 module started).  Finishing out that build, I found that the 
2.6.27.7 stock kernel would boot and NOT resync the array.  That proves, 
to me, something is wrong with the overall Mandriva 2009.0 system (as 
well as 2008.1 which also fails miserably).

<sigh>  I got nailed on this one!  I updated the machine about two 
months before I did a shutdown prior to moving to my new home.  Just 
before I left the old place, I booted this box to do some financial 
stuff and found it resync-ing the raid5 array...

Thank you again.

Just a follow-up.  Add the internal bitmap to the arrays has cured the 
problem (mdadm --grow --bitmap=internal <mdX>).  This is definitely a 
feature that I will consider adding to the existing raid5 arrays that I 
maintain out in the wild.

Is this a non-destructive thing to do to a working array?  Grow it with 
adding the bitmap?  I did not make each member partition of the arrays 
consume the remainder of the drives, but left 50..150 blocks unused in 
them (found various 250Meg drives do not all have the same block counts 
between manufacturers).

Regards,

TomW

-- 
Tom Walsh - WN3L - Embedded Systems Consultant
http://openhardware.net http://cyberiansoftware.com http://openzipit.org
"Windows? No thanks, I have work to do..."
----------------------------------------------------