From mboxrd@z Thu Jan  1 00:00:00 1970
Message-ID: <51866f86da16fb592ccf52a3223ef4de@collyer.net>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] some shell scripts
Date: Thu, 14 Oct 2004 17:25:45 -0700
From: geoff@collyer.net
In-Reply-To: <416E9B72.9030400@9fs.org>
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Topicbox-Message-UUID: ee72e556-eacd-11e9-9e20-41e7f4b1d025

One could start lnfs early in cpurc and termrc, have it post a /srv
file, and have /lib/namespace* mount that /srv file.  I think that for
the sort of thing you described (with Beagle), creations and removals
are of primary interest, though one could watch for writes too.
Keeping a database or log of such changes (perhaps served from the
watcher file server, permitting some in-memory caching) is vastly
cheaper than statting a potentially huge set of files every second
(though the difference would be even bigger on (l)unix, with the
directory entries separated from the i-nodes).  One should be able to
just note the new file name, size and perhaps mtime of each write, so
the volume of data should be tolerable.  If you're building indices, a
watcher client would probably want to wait for a given file to be
closed and perhaps wait a little longer in case it's rewritten, then
index it.  Maybe index every changed file each hour, or something of
the sort.

I don't think the need kernel mods for this.