Re: [PATCH 1/3] libnsm.a: sm-notify sometimes ignores monitored hosts

[Steve Dickson <SteveD@redhat.com>]

Hey...

On 12/13/2010 12:08 PM, Chuck Lever wrote:

>>>  */
>>> @@ -863,10 +879,15 @@ nsm_load_host(const char *directory, const char *filename, nsm_populate_t func)
>>> 	if (path == NULL)
>>> 		goto out_err;
>>>
>>> -	if (stat(path, &stb) == -1) {
>>> +	if (lstat(path, &stb) == -1) {
>>> 		xlog(L_ERROR, "Failed to stat %s: %m", path);
>>> 		goto out_freepath;
>>> 	}
>>> +	if (!S_ISREG(stb.st_mode)) {
>>> +		xlog(D_GENERAL, "Skipping non-regular file %s",
>>> +				path);
>> Question, why do we care non-regular files are being ignored?
> 
> We probably want to report anything unexpected in 
> the /var/lib/nfs/statd/sm{,.bak} directories.
> 
>> I understand logging the lstat() error but logging statements like
>> "ignoring this" or "not doing that" just make the debug output a 
>> bit too noisy IMHO... 
> 
> My expectation is that under normal circumstances this message 
> would never fire.  
Correct the only way they will be shown is with the "-F -d" flags... 

> statd shouldn't put anything in that directory 
> that isn't a regular file.  If there's something else in there, 
> we should report it.  Also, if statd (or something else) is broken 
> and the code thinks the object isn't a regular file, then this 
> message would point to what is wrong.
To put this in context, here are the message that come up with
the patch applied... I added a symlink to /var/lib/nfs/statd/sm/statd
to generate both messages... 

statd: Version 1.2.3 starting
statd: Flags: No-Daemon Log-STDERR TI-RPC 
sm-notify: Version 1.2.3 starting
sm-notify: Already notifying clients; Exiting!
statd: Skipping dot file ..
statd: Skipping non-regular file /var/lib/nfs/statd/sm/statd
statd: Skipping dot file .

I think we both agree the "Skipping dot file" simply not
needed... The "Skipping non-regular" message... well I have 
to say it really does not have much meaning... either.. IMHO...

But I do agree, it probably will *never* seen... 

steved.