RAID arrays not reconstructing in 2.6
I saw this a couple of weeks ago in a 2.6.0-test kernel, and today in 2.6.2-rc3. When I have to hit the hard reset button on my desktop, whose root filesystem is RAID5 on /dev/md0, it comes back up cleanly - no reconstruction.
Have we gotten a whole lot more enthusiastic about marking superblocks clean lately, or should I be worried? Obviously this always used to trigger reconstruction, until recently.
Signature · 2 lines
Daniel Jacobowitz MontaVista Software Debian GNU/Linux Developer
Quoted 4 lines · On Sunday February 1, [email protected] wrote:
I saw this a couple of weeks ago in a 2.6.0-test kernel, and today in 2.6.2-rc3. When I have to hit the hard reset button on my desktop, whose root filesystem is RAID5 on
/dev/md0, it comes back up cleanly - no reconstruction.
Cool, isn't it!
Quoted 4 lines
Have we gotten a whole lot more enthusiastic about marking superblocks clean lately, or should I be worried? Obviously this always used to trigger reconstruction, until recently.
Yes. Lots more enthusiastic.
If there is no write activity for 20msec, we mark the superblock clean and write it out, and are careful to write out a dirty superblock before allowing another write to complete.
NeilBrown
Quoted 4 lines
Signature · 3 lines
-- Daniel Jacobowitz MontaVista Software Debian GNU/Linux Developer
Quoted 9 lines · Neil Brown
Quoted 4 lines · On Sunday February 1, [email protected] wrote:
I saw this a couple of weeks ago in a 2.6.0-test kernel, and today in 2.6.2-rc3. When I have to hit the hard reset button on my desktop, whose root filesystem is RAID5 on
/dev/md0, it comes back up cleanly - no reconstruction.Cool, isn't it!
Quoted 1 line
Quoted 8 lines
Quoted 3 lines
Have we gotten a whole lot more enthusiastic about marking superblocks clean lately, or should I be worried? Obviously this always used to trigger reconstruction, until recently.
Yes. Lots more enthusiastic.
If there is no write activity for 20msec, we mark the superblock clean and write it out, and are careful to write out a dirty superblock before allowing another write to complete.
Wow, thanks! That's really awesome, considering how long reconstruction takes this system (about an hour of completely useless otherwise).
Signature · 2 lines
Daniel Jacobowitz MontaVista Software Debian GNU/Linux Developer