Re: [RFC/PATCH] inotify -- a dnotify replacement

[John McCutchan]

On Tue, 2004-05-11 at 07:52, nf wrote:
> On Tue, 2004-05-11 at 04:47, Chris Wedgwood wrote: 
> > On Mon, May 10, 2004 at 06:17:40PM -0400, John McCutchan wrote:
> > 
> > > According to everyone who uses dnotify it is.
> > 
> > I don't buy that.  I have used dnotify and signals where not an issue.
> > Why is this an issue for others?
> 
> I believe the worst thing about dnotify is the "umount blocking"
> behaviour. It drives me crazy since i use the Linux desktop. So if this
> is the "Year of the linux desktop", please please switch OFF dnotify
> until it does not open files for monitoring anymore! Or until "inotify"
> works.
> 
> Btw, i have written a little tool to assist people with the
> umount-problem and collected some links
> http://www.scheinwelt.at/~norbertf/wbumount/. 

That will be fixed with inotify.

> 
> > > The idea is to encourage use of a user-space daemon that will
> > > multiplex all requests, so if 5 people want to watch /somedir the
> > > daemon will only use one watcher in the kernel. The number might be
> > > too low, but its easily upped.
> > 
> > If you are to use a daemon for this, why no use dnotify?
> 
> I don't understand, why the author of inotify wants to force people to
> use user-space daemons like fam (which requires xinetd, ...)? Does using
> a daemon for multiplexing really add efficiency? Is it really worth
> adding all the complexity of things like "fam" to the system, just
> because more than one application monitor the same directory once in a
> while? I doubt it.
> 

I doesn't have to be fam. I am not sure if having a daemon is the best
way to go either, but I think that having a daemon (not necessarily fam)
could be beneficial. Beyond multiplexing you could keep a buffer of
events and try and combine some events before sending them to the apps.
But if it turns out that a daemon is not the best way, we can adjust the
limits set on the driver and everyone can just use it directly. So relax
I am not trying to force anything, it was just an idea.

John
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/