[Sheepdog] Few things

[MORITA Kazutaka]

Hi,

At Sun, 22 May 2011 13:27:15 +0200,
krimson wrote:
> 
> Hello,
> 
> First of all, thanks for working on sheepdog, it is a project that looks 
> very promising. I've been following this list for a while now and 
> finally started testing sheepdog yesterday. I have a couple of 
> questions/remarks:
> 
> 1) if I do not specify -l LOG_ERR to the sheep daemon it creates a lot 
> of logging to sheep.log and eats up a lot of CPU in the process, making 
> operations like qemu-img convert and create very very slow. It seems to 
> me LOG_ERR should be the default instead of LOG_INFO ?

I think the right way is using more appropriate log levels for log
messages.  Currently, we output too many logs with a high log level.

> 
> 2) if the sheep store is unavailable (for example the sheepdog device 
> was not mounted), the sheep daemon tends to hang and consume 100% CPU 

What are in the sheep store before you mounted the sheep device?
Unfortunately, I couldn't reproduce the problem.  I tried

 $ sheep -f /dir       # /dir does not exist
 $ sheep -f /          # I don't have a permission to write /
 $ sheep -f /file      # /file is a file

In any case, the sheep program exits just after I ran the command, and
doesn't consume any CPU resource.  Could you explain more details to
produce the problem?


> (possibly also to the logging issue). If it gets confused in this way, I 
> have to rm -rf /sheep/* before I can restart the sheep daemon and have 
> it join the cluster again. Perhaps add a little more robustness to this 
> scenario :)
> 
> 3) Is the trac site the proper place to report issues ? If so, it would 
> be nice if you could add the GIT repository to it if possible, so we can 
> browse and see the changes in Timeline :) There seems to be a repository 
> attached right now but the last change was 7 weeks ago which doesn't 
> seem right.

The trac on sourceforge.net doesn't support a git repository, so I
manually synchronize the sheepdog git tree to a svn tree...

I've synchronized just now, but I think we should do better:
 - synchronize automatically,
 - remove a source code tree from the Trac to avoid confusion,
 - move to another site which supports a git tree, or
 - move to another project management system.

Any ideas?


Thanks,

Kazutaka