Date: Thu, 2 Apr 2009 18:50:24 +0200
From: Jan Kara <jack@suse.cz>
To: Alexander Larsson <alexl@redhat.com>
Cc: eparis@redhat.com, linux-kernel@vger.kernel.org
Subject: Re: Issues with using fanotify for a filesystem indexer
Message-ID: <20090402165023.GG3010@duck.suse.cz>
References: <1238158043.23703.20.camel@fatty> <20090402145457.GA17275@atrey.karlin.mff.cuni.cz> <1238689744.5704.1.camel@localhost.localdomain>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1238689744.5704.1.camel@localhost.localdomain>
User-Agent: Mutt/1.5.17 (2007-11-01)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2335
Lines: 47

On Thu 02-04-09 18:29:04, Alexander Larsson wrote:
> On Thu, 2009-04-02 at 16:54 +0200, Jan Kara wrote:
> 
> >   Some time ago I was trying to solve a similar problem and I've come up
> > with a solution which I've called recursive mtime. The general idea is
> > that with each directory, kernel additionally keeps a flag and a
> > timestamp. When a directory is modified, we do:
> >   dir = changed dir;
> >   while dir has flag set do
> >     update timestamp to current time
> >     clear flag
> >     dir = parent dir
> > 
> >   When a file is modified, you just start with a parent directory of
> > that file. With this scheme, you are able to find reasonably quickly
> > (without looking at unchanged directories) what has changed since
> > you've looked last time (you look for timestamps newer than the time
> > when you started last scan and you set flags as you go). Also the scheme
> > is quite cheap to maintain and has no problems with overflowing event
> > queues etc. (observe that the scheme works perfectly fine for several
> > independent scanners in parallel). As a bonus, if you store the flag +
> > timestamp persistently on disk, you can use this scheme to speedup things
> > like rsync.
> >   What gets nasty (but solvable) are hardlinks and bind mounts. I was
> > writing a library to handle these last summer but then had to work on
> > something else and didn't get back to it yet.
> 
> Another potential issue with this is that every change bubbles up to the
> top, modifying the recursive mtime of that. This will become very
> contented, and may imply a partial serialization of fs activity, which
> is kinda costly.
  Not every change - only the first change bubbles to the top, clearing the
flag on its way. Then next change stops bubbling up as soon as it reaches
a directory with the flag cleared. So no contention happen - we update flag
+ timestamp only at most once per scan of the directory by indexer (or
someone else interested in recursive mtime) => once per a few minutes on
average system.

									Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/