[sheepdog] filesystem errors when using sheepdog

Vasiliy Tolstov v.tolstov at selfip.ru
Fri Jul 17 09:13:52 CEST 2015


2015-07-17 8:58 GMT+03:00 Hitoshi Mitake <mitake.hitoshi at gmail.com>:
> Vasiliy,
>
> https://github.com/sheepdog/qemu/tree/serialize-v3
>
> I updated the patch in the above branch. Could you test it?

Thanks! I'm test it and send results.

>
> BTW, could you describe your way of reproducing the bug? We are also
> trying to reproduce but reproducibility is not so good (we are using
> fio). It seems that your testing method has a high reproducibility. If
> you can share it, it is a great help.

My test is very simple - i have 3.18.3 vanilla kernel with compiled in
virtio drivers and custom initrd that contains only one binary -
binary written in golang. I have prebuilded images (builded from
packer - raw qemu image) that compressed by bgzf (parallel gzip
encoder/decoder). After vm starts it fetches image from http server on
the fly decompress and write to disc. After write successed  it send
sync to fs and restart vm.
So i have this works fine with qemu and plain lvm, on unpatched qemu
2.3.0 with limiting bandwidth or installing only on vm at the same
time all works fine, but if vm count more than 3-5 at the same time
for one node i have errors that i send to list.
If this version of patch does not helps i'm tr to rewrite vm installer
and share it in boxed variant and send instructions how to run it in
qemu.


-- 
Vasiliy Tolstov,
e-mail: v.tolstov at selfip.ru


More information about the sheepdog mailing list