[sheepdog] [PATCH] add INFO level operation logging on each node.

[MORITA Kazutaka]

At Wed, 22 Jan 2014 17:41:18 +0800,
Liu Yuan wrote:
> 
> On Wed, Jan 22, 2014 at 06:18:52PM +0900, Hitoshi Mitake wrote:
> > At Tue, 21 Jan 2014 20:38:07 +0900,
> > Matsuo Yoshinori wrote:
> > > 
> > > I think the both logging function are needed because of the reasons below.
> > > 
> > > Reason:
> > > 1. The dog and the sheep are another process.
> > > Even if dog is working good but sheep process may have some trouble.
> > > Then we need to track on the sheep process, too.
> > > 
> > > 2. The reason that we want a logging function on sheep process.
> > > The sheep processes are always in the cluster, but dog are not.
> > > We may not be able to get the evidence log,
> > > if dog command was issued from unknown or operators temporary host.
> > > 
> > > 3. The reason that we want a logging function on dog command.
> > > Sheep can't log dog command level directry but dog can log more readable
> > > log with all command options.
> > > The operator wants to see dog command level evidence first, when checking
> > > what is done to cluster.

What do you mean by "dog command level"?


> > > 
> > > I would like to know Hitoshi's opinion, too.
> > 
> > I agree with Matsuo-san's opinion. dog and sheep can be executed on
> > different hosts, so log for evdence should be written by both of
> > them.
> > 
> > In addition, we are building a product, not a service. We need to
> > analyze problems based on logs produced by users' deployments when
> > they are in trouble. For doing smooth trouble shooting, we need logs
> > as much as possible. Because we wouldn't be able to access to users'
> > deployments directly.
> > 
> > For expanding use case of sheepdog, this logging enhancement would be
> > helpful.
> 
> I am doubtful that these logs can really help debug and if you are selling
> service, you should help people who buys your service to deploy sheepdog
> correctly. (Deployment can be done automatically by puppet or chef, so what we
> need is actually provide mechanisms at best. One mechanism for one problem)
> 
> I personally disagree duplicate the functionality both in dog and sheep, but if
> all other developers are not against it. So I want to hear kazutaka's opinion
> on this patch.

IMHO, I don't really understand why we have to log dog operations in
the dog program.  Why can't we do it outside of dog?  I mean it looks
enough to create a wrapper script like as follows.

 echo begin <command>
 dog <command>
 echo end <command>

I'm not against adding more log outputs in sheep, but this patch
doesn't look good to me.  I'll add some comments later.

Thanks,

Kazutaka