[sheepdog-users] cache flush: all or nothing

[Valerio Pachera]

2013/6/6 Liu Yuan <namei.unix at gmail.com>:
> If the VM is doing nothing and shutdown takes a long time, I think there
> is something wrong.

On the test cluster:
- a quest with a vdi formatted with ext3
- wrote 4G of data and waited more than 30 minutes.
- Flushed the cache


root at test004:~# time collie vdi cache flush squeeze

real    0m31.903s
user    0m0.000s
sys     0m0.000s

Then I powered off the guest.
It took 1':28'' since I got "It will now halt".
If I start and shutdown the guest without writing data, it takes 1
second or less.

This is not the only test I've done but I got non repeatable results.
I may report that later to avoid confusion.

root at test004:~# ps aux | grep sheep
root      2738  3.0  4.7 343044 91180 ?        Sl   12:07   2:50 sheep
-w size=20000 /sheep/sda /sheep/sdb
root      2739  0.0  0.0  83408   468 ?        Ss   12:07   0:00 sheep
-w size=20000 /sheep/sda /sheep/sdb

qemu-system-x86_64 -name squeeze -enable-kvm -drive
file=sheepdog:squeeze,cache=writeback -drive
file=sheepdog:data,cache=writeback -m 512 -smp 2 ...

root at test004:~# sheep -v
Sheepdog daemon version 0.6.0

QEMU emulator version 1.5.50