[sheepdog-users] Time to get a node info or md info

[Liu Yuan]

On Fri, Aug 09, 2013 at 06:32:47PM +0200, Valerio Pachera wrote:
> 2013/8/9 Valerio Pachera <sirio81 at gmail.com>:
> > May it be possible/clever to retrieve the info from df ?
> 
> I guess 'df' uses some c function that collie may use too.
> Sure I'm not aware of many (all) back ground stuff, but just to give
> you an idea of what I would like to see (in term of speed).
> 
>  time parallel-ssh -i -h pssh.conf 'df -h | grep /mnt/sheep'
> [1] 18:28:47 [SUCCESS] sheepdog001
> /dev/mapper/vg00-sheepdsk01                             167G  137G
> 30G  83% /mnt/sheep/dsk01
> /dev/sdb1                                               466G  329G
> 137G  71% /mnt/sheep/dsk02
> /dev/sdc1                                               1,9T  1,2T
> 709G  62% /mnt/sheep/dsk03
> [2] 18:28:47 [SUCCESS] sheepdog004
> /dev/mapper/vg01-sheepdsk01   50G   50M     50G   1% /mnt/sheep/meta
> /dev/sdc1                    466G  216G    251G  47% /mnt/sheep/dsk03
> /dev/sdd1                    1,9T  1,4T    449G  76% /mnt/sheep/dsk04
> [3] 18:28:47 [SUCCESS] sheepdog003
> /dev/sda6       170G  154G     16G  91% /mnt/sheep/dsk01
> /dev/sdb1       2,8T  1,9T    934G  67% /mnt/sheep/dsk02
> [4] 18:28:47 [SUCCESS] sheepdog002
> /dev/mapper/vg00-sheepdsk01                             167G   85G
> 82G  51% /mnt/sheep/dsk01
> /dev/sdb1                                               466G  300G
> 167G  65% /mnt/sheep/dsk02
> /dev/sdc1                                               1,9T  1,2T
> 668G  65% /mnt/sheep/dsk03
> 
> This applicable only if sheep runs on dedicated devices.
> 
> 
> real    0m0.287s
> user    0m0.056s
> sys     0m0.036s

The long wait time doesn't attribute to how sheepdog get the disk usage, but how
busy sheep daemon is. In the recovery, sheep main loop is saturated by recovery
requests and the 'md info' get kind of starved.

We'll optimize the main loop after v0.7.0, probably this problem can be mitigated.

Thanks
Yuan