mount recovery taking quite long time

mount recovery taking quite long time

[Arkadiusz Miśkiewicz]

3.0.17 kernel, 800GB fs, fresh boot, mount of /home at boot (so no daemons 
running yet),  and :

[   17.768459] XFS (md3): Starting recovery (logdev: internal)
[  872.427061] XFS (md3): Ending recovery (logdev: internal)

That mount was after (unfortunate) machine reset while it was doing many 
things.

The question is - what could take so long in here?
-- 
Arkadiusz Miśkiewicz        PLD/Linux Team
arekm / maven.pl            http://ftp.pld-linux.org/

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: mount recovery taking quite long time

[Dave Chinner]

On Wed, Feb 15, 2012 at 12:25:13PM +0100, Arkadiusz Miśkiewicz wrote:
> 
> 3.0.17 kernel, 800GB fs, fresh boot, mount of /home at boot (so no daemons 
> running yet),  and :
> 
> [   17.768459] XFS (md3): Starting recovery (logdev: internal)
> [  872.427061] XFS (md3): Ending recovery (logdev: internal)
> 
> That mount was after (unfortunate) machine reset while it was doing many 
> things.
> 
> The question is - what could take so long in here?

How big is the log? It's entirely possible that your log is full of
modified inodes and so it's having to read them all in first. A
modified inode only takes a couple of hundred bytes in the log, so
even if you have a 100MB log, you can still have a couple of million
dirty inodes in the log that have to be read and replayed.

This has been known for some time, but the worse case (around
15 minutes - what you've seen here - is most I've seen or heard
reported) is relatively rare so there hasn't been any urgency to
solve this:

http://xfs.org/index.php/Improving_Metadata_Performance_By_Reducing_Journal_Overhead#Reducing_Recovery_Time

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs