[sheepdog-users] Undestand disk conversion

Valerio Pachera sirio81 at gmail.com
Wed Dec 19 16:54:50 CET 2012


Hi, I just converted a raw disk of 30G on a newly formated cluster (4
nodes 2 copies).

I run
  qemu-img convert -f raw win7.raw sheepdog:win7
on the 4th node that is using an ssd disk.

Conversion ha been really fast: a bit more than 4 minutes.

What I thought to see during conversion:
- i/o activity on disk and network on all nodes
- collie node info increasing the percentage on all nodes

What I saw:
- the other 3 nodes, no i/o activity
- no increasing of use percentage
  collie node info  (also once terminated the process)
  Id      Size    Used    Use%
   0      216 GB  4.0 MB    0%
   1      216 GB  4.0 MB    0%
   2      216 GB  0.0 MB    0%
   3      112 GB  0.0 MB    0%
  Total   760 GB  8.0 MB    0%
- loopback reading and wrting 900M/s
- sda busy 90% at reading (the disk where win7.raw is stored)
- my ssd disk 50% busy at writing

I was thinking that the convert process stores all the chunks in the
node doing running qemu-img, and later it was going to distribute the
chunks to the other nodes.
I also thought it maybe that was visible by 'collie node recovery',
but nothing of that has happened.

collie node list
M   Id   Host:Port         V-Nodes       Zone
-    0   192.168.2.41:7000      73  688040128
-    1   192.168.2.42:7000      73  704817344
-    2   192.168.2.43:7000      73  721594560
-    3   192.168.2.215:7000     38 -687691584

collie vdi list
  Name        Id    Size    Used  Shared    Creation time   VDI id  Copies  Tag
  win7         1   30 GB   30 GB  0.0 MB 2012-12-19 16:00   605eca     2

I run a vm with sheepdog:win7, then I saw 'collie node info'
increasing the use percentage.
I had to wait something like 15/20 minutes before it finished the
processes and I notice that looking atop.

collie node info
Id      Size    Used    Use%
 0      216 GB  18 GB     8%
 1      216 GB  17 GB     7%
 2      216 GB  16 GB     7%
 3      112 GB  9.6 GB    8%
Total   760 GB  60 GB     7%
Total virtual image size        30 GB

Is there a way to know the state of the conversion process? (Something
like 'collie node recovery).
Vnc of the guest is like frozen till the conversion is done.
Is there a way to force sheepdog to complete the disk conversion
without running a guest?



More information about the sheepdog-users mailing list