On Sun, Apr 17, 2005 at 10:49:14AM -0700, Tim Moore wrote: Hi, > The recovery daemon adjusts reconstruction speed dynamically according to > available system resources. > Disk I/O is somewhat slower but works just fine. You don't have to wait. So I don't have to wait to take the disk out, as the recovery will continue with embedded disk battery and wireless bus connection? How cool... ;-) Well... more seriously, I can't believe this question doesn't raise any interest, even if it seems like it does not. :-( Does everyone really type cat /proc/mdstat from time to time?? How clumsy... I just want to chat about the best way to add a backend for this kind of feature, so we could implement that properly... (and yes, that is definitely _nedded_ if you want to do things right) Herve > Hervé Eychenne wrote: > > Hi, > > > >Suppose I'm waiting for a recovery to be completed, and want to run a > >command afterwards (halt, send a mail, or anything else...). > >The most practiacl way I can see is to check /proc/mdstat. > > > >But what if I want to do that automatically (without bothering looking > >at it manually from time to time)? > >For example, one could do: > ># while cat /proc/mdstat | grep recovery > /dev/null ; do sleep 5 ; done > > > >But that's quite ugly, as: > >- it's an active polling, and it is time consuming (even if slightly) > >- it may even be unreliable, as I guess one cannot ensure that /proc/mdstat > > will print the "recovery" string during the (very short, but well...) > > transition between two partitions to recover > > > >I think that a passive wait would be much better instead. > >And ideally, we should have a simple and efficient way to let a program > >know if a device is in a clean state (or being recovered), and another > >that would wait until the device is clean (recovery finished). > > > >So, the while loop could be replaced by something like > > mdadm --recovery-wait (for example) > >which would exit only when all pending recoveries have finished, and > >let the script continue. > >That would be much practical, reliable, and cleaner than a loop, don't you > >think? > > > >How this could be achieved is another question... probably the best > >would be that userspace can select on a file descriptor, or something > >like that (netlink device?) > >What do you think? > > > > Hervé Hervé -- _ (°= Hervé Eychenne //) Homepage: http://www.eychenne.org/ v_/_ WallFire project: http://www.wallfire.org/ - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
