[sheepdog] [PATCH v3 3/7] sheep: teach journal file to handle epoch and config write

Hitoshi Mitake mitake.hitoshi at gmail.com
Fri Apr 5 11:23:02 CEST 2013

At Fri, 05 Apr 2013 17:15:16 +0800,
Liu Yuan wrote:
> On 04/05/2013 05:06 PM, Hitoshi Mitake wrote:
> > Simply causing panic() makes the patchset smaller. I feel the current
> > behavior is too dangerous because users have no way to know disabling
> > of journals until they read the log messages.
> Actually, from the feedback of our production system, users are more
> sensitive to panic. Panic means VM resource is suddenly unusable and a
> big possibility to break the VM's FS. This will always be last resort to
> consider. Sheepdog plays a more vital role than hyper visor in cloud
> computing environment because it deals with data directly. Disable
> journaling is not as dangerous than panic out. Or what exactly danger
> you think of about disabling journaling?

I see. I thought disabling journaling is too dangerous because the
behavior cannot satisfy users' requirements which is given when sheep
process is invoked.

But as you say, panic would also be dangerous in production
environments. I missed the perspective. Let's implement the option for
controlling the behavior later.


More information about the sheepdog mailing list