[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [dm-devel] [PATCH] Don't print failure messages for callouts by default



Hi Hannes,

On Fri, 02 May 2008 08:55:11 +0200, Hannes Reinecke wrote:
> >> @@ -642,7 +642,7 @@ get_prio (struct path * pp)
> >>  	}
> >>  	pp->priority = prio_getprio(pp->prio, pp);
> >>  	if (pp->priority < 0) {
> >> -		condlog(0, "%s: %s prio error", pp->dev, prio_name(pp->prio));
> >> +		condlog(3, "%s: %s prio error", pp->dev, prio_name(pp->prio));
> >>  		pp->priority = PRIO_UNDEF;
> >>  		return 1;
> >>  	}
> > 
> > When prio_getprio() fails, that path is not used for multipath.
> > I think that it is a kind of serious error and the verbosity should
> > be "2" at least.
> > 
> > Why do you need to suppress this message by default?
>
> Because the prioritizer also will fail if a path failed. In this case
> you get quite some errors like 'prio_alua failed' on a failed path,
> inducing you to check if the prio_alua program might be wrong.
> In fact, everything is alright and working as expected, only
> that a path has failed.
> So this error message might a well be suppressed.
 
OK, agreed on this.


> > Hi Hannes,
> > 
> > On Wed, 30 Apr 2008 11:04:26 +0200, hare suse de (Hannes Reinecke) wrote:
> >> Calling 'multipath -ll' on devices with paths results in
> >> lots of error messages; they really should be suppressed
> >> for the normal output. The user can always have them printed
> >> out by increasing verbosity.
> > 
> > In general, serious error messages should be printed by default.
> > Otherwise, the users (even developers) may overlook serious errors.
> > 
> Agreed in principle. However, most of these error messages are secondary
> messages (ie the are generated after the principal cause was reported)
> so they do not carry any additional information to the 'normal' user.
> 
> Case in point here is for a path failure: this is multipathing, after all,
> the whole point of which is that we _can_ handle path failure seamlessly.
> So throwing error messages for a perfectly normal flow of operation is
> really confusing.
> 
> > Please see the detailed comments below.
> >  
> >> @@ -32,7 +33,7 @@ int execute_program(char *path, char *value, int len)
> >>  	int retval;
> >>  	int count;
> >>  	int status;
> >> -	int fds[2];
> >> +	int fds[2], null_fd;
> >>  	pid_t pid;
> >>  	char *pos;
> >>  	char arg[PROGRAM_SIZE];
> >> @@ -75,7 +76,16 @@ int execute_program(char *path, char *value, int len)
> >>  		close(STDOUT_FILENO);
> >>  
> >>  		/* dup write side of pipe to STDOUT */
> >> -		dup(fds[1]);
> >> +		if (dup(fds[1]) < 0)
> >> +			return -1;
> >> +
> >> +		/* Ignore writes to stderr */
> >> +		null_fd = open("/dev/null", O_WRONLY);
> >> +		if (null_fd > 0) {
> >> +			close(STDERR_FILENO);
> >> +			dup(null_fd);
> >> +			close(null_fd);
> >> +		}
> >>  
> >>  		retval = execv(argv[0], argv);
> >>  
> > 
> > This looks discarding all error messages from callouts anyway,
> > even if we want to get them.
> > Can you use the verbosity setting here, too?
> > 
> Yes, we could; no problem.

Thanks.
This part should print primary error messages which include much
information to analyze the cause of the error.

 
> >> @@ -663,7 +663,7 @@ get_uid (struct path * pp)
> >>  		condlog(0, "error formatting uid callout command");
> >>  		memset(pp->wwid, 0, WWID_SIZE);
> >>  	} else if (execute_program(buff, pp->wwid, WWID_SIZE)) {
> >> -		condlog(0, "error calling out %s", buff);
> >> +		condlog(3, "error calling out %s", buff);
> >>  		memset(pp->wwid, 0, WWID_SIZE);
> >>  		return 1;
> >>  	}
> > 
> > ditto.
> > 
> Cheers,

OK.  When the primary messages above are printed correctly,
this secondary message can be suppressed by default.

Thanks,
Kiyoshi Ueda


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]