bcache, layered block devices and unclean shutdowns

bcache, layered block devices and unclean shutdowns

[Kingsly John]

Hi!

In the archives I found a thread from a few months ago where it was
recommended that it's best to use bcache on the RAID layer.(rather than the
disks themselves or LVM)

Wouldn't this affect bcache's ability to recover from an unclean shutdown?
ie, if the raid array itself can't be brought up, bcache wouldn't be able to
write caches to disk? (Or is that not a possibility)

Kingsly

-- 

Re: bcache, layered block devices and unclean shutdowns

[Kent Overstreet]

On Thu, Oct 04, 2012 at 04:06:25PM +0530, Kingsly John wrote:
> Hi!
> 
> In the archives I found a thread from a few months ago where it was
> recommended that it's best to use bcache on the RAID layer.(rather than the
> disks themselves or LVM)
> 
> Wouldn't this affect bcache's ability to recover from an unclean shutdown?
> ie, if the raid array itself can't be brought up, bcache wouldn't be able to
> write caches to disk? (Or is that not a possibility)

Not sure what you're asking...

Re: bcache, layered block devices and unclean shutdowns

[Kingsly John]

+++ Kent Overstreet [2012-10-04 12:51:30]:

> On Thu, Oct 04, 2012 at 04:06:25PM +0530, Kingsly John wrote:
> > Hi!
> > 
> > In the archives I found a thread from a few months ago where it was
> > recommended that it's best to use bcache on the RAID layer.(rather than the
> > disks themselves or LVM)
> > 
> > Wouldn't this affect bcache's ability to recover from an unclean shutdown?
> > ie, if the raid array itself can't be brought up, bcache wouldn't be able to
> > write caches to disk? (Or is that not a possibility)
> 
> Not sure what you're asking...

If /dev/md0 is the backing device, in the event of an unclean shutdown during
a write it would be dirty(at the raid level) and there would be a resync
etc. 

If the individual disks that make make up /dev/md0 were backing devices
instead, wouldn't bcache ensure that /dev/md0 would never end up dirty when
recovering from an unclean shutdown and eliminate the need for a forced
resync?

Is one better than the other in terms of maintaining data integrity or would
they both be equally reliable?

Kingsly

--