MIME-Version: 1.0
In-Reply-To: <rt-3.8.6-60453-1284580962-1742.22096-5-0@isc.org>
References: <RT-Ticket-22096@isc.org> <38C0BFD50092414E900DEE15F8976FE0@sb.litts.net> <rt-3.8.6-60453-1284580962-1742.22096-5-0@isc.org>
Message-ID: <AA666FB582F2412991E8AE7F9568631A@sb.litts.net>
Content-Type: text/plain; charset="utf-8"
Subject: RE: [ISC-Bugs #22096] FW: Statistics channel patch 
Thread-Index: ActVEPsNHiXTYchfS7ClKZo1M9bHzAAGRgRA
Date: Wed, 15 Sep 2010 20:17:28 -0400
To: <bind-suggest@isc.org>
Content-Transfer-Encoding: 7bit
From: "Timothe Litt" <litt@acm.org>
RT-Message-ID: <rt-3.8.6-60449-1284596270-1934.22096-0-0@isc.org>
Content-Length: 4750

Michael, 

Thanks for the reply.  

I'm not too fussy about exactly what things look like, as long as it can
function.  Here are my 3 cents:

I think config, stats, status are orthogonal selectors to the data's
structure - so why not put them in the query string?

When one knows what one needs, it's best not to have to make multiple
requests.  (.g. /views to get the view list, then /views/eachview/zones to
get the zone list, then /views/eachview/zones/zone/eachzone to get the zone
type...)

For my present use, I build a dropdown for actively served, user
(non-default) zones.  This happens fairly frequently.  So it's important to
keep it efficient.  

I don't want to make more than one query to the server to get this
configuration data.
So maybe something like:

   /views/*/zones?view.zone.type=user&DataClass=config

would work.  The idea is that the query could be any xml path=desired value,
and DataClass is a metavariable.  Alternatively, if you want to think about
the Data class as a part of the structure, I'd put in front, not after.
This because the rest of the path is going from the most general to the most
specific.  And /config/views/*/zones/* ... means "just config data at every
level", not just at the last one...

Of course, my litmus test is whether I can issue one request and get what I
need for my application.

Perhaps an example to test your design against would help.  The outline of
my code is:
	Fetch config
	checksum
	if matches cache, use cache (don't even parse the string)
	else update cache & use result.

Stripping out error handling and a bunch of other details (for clarity), my
current code looks like:

my $config = LWP::Simple::get( "http://$masterNs:8053/config" );
my $checksum = Digest::SHA1::sha1($config);
my $cfgname = dirname(Cwd::realpath( $0 )) . '/DnsConfig.db';
My( $cfg, %zones, %validzone );
eval { cfg = Storable::lock_retrieve($cfgname); };
if( $cfg && $cfg->[0] == 1 && $cfg->[1] eq $checksum ) {
	%zones = @{$cfg->[2]};
	%validzone = @{$cfg->[3]};
        ...
} else {
    my $xs = XML::Simple->new();
    # Extract views (because it speeds up parsing)
    $config =~ m/(<views>.*?<\/views>)/ms or die "Can't find views\n";;
    my $views = $1;
    $views = $xs->parse_string( $views );
    # Build %zones from server's active view/zone list
    for my $view (keys %{$views->{view}}) {
	my $zones = $views->{view}{$view}{zones}{zone};
	my( @fwd, @rev );
	for my $z (keys %$zones) {
	    next if( $zones->{$z}{type} ne 'user' );
	    next if( lc( $zones->{$z}{rdataclass} ) ne 'in' );
	    next unless( $z =~ m!^(.*)/IN/$view$! );
	    $z = lc( $1 );
	    if( $z =~ /^(.*)\.in-addr.arpa$/ ) {
		  my @a = reverse( split( m/\./, $1 ) );
		  my $a = join( '.', @a );
		  # Split out [0] private and [1] public IP zones
		  # note that is_privateip returns a classification code; !
reduces to 0 or 1
		  push @{$rev[!is_privateip($a . scalar('.0' x (4-@a) ))]},
[ $a, "$z/$view" ];
		  $validzone{"$z./$view"} = $a;
		} else {
		  # Forward zones
		  push @fwd, [ $z, "$z/$view" ];
		  $validzone{"$z./$view"} = $z;
	    }
	    next unless( @fwd || @rev );        # Skip empty views
	    $zones{$view}{forward} = [ @fwd ] if( @fwd );
	    $zones{$view}{reverse}{private} = sortip @{$rev[0]} if( $rev[0]
);
	    $zones{$view}{reverse}{public} = sortip @{$rev[1]} if( $rev[1]
);
	}
    }
    Storable::lock_store( [ 1, 		# [0] Version
				    $checksum,	# [1] Checksum of received
network data
				    [%zones],	# [2]
				    [%validzone],	# [3]
					...
			         ], $cfgname );
}
---------------------------------------------------------
This communication may not represent my employer's views,
if any, on the matters discussed. 
 
-----Original Message-----
From: Michael Graff via RT [mailto:bind-suggest@isc.org] 
Sent: Wednesday, September 15, 2010 16:03
To: litt@acm.org
Subject: [ISC-Bugs #22096] FW: Statistics channel patch 

Thank you for your feedback on the XML channel.

>From a design point of view, I intentionally did not split the data apart
but
instead returned the whole thing. This was intentional until we could make
it
into a more RESTful style.

It is probably time to actually take on the work of splititng the data into
chunks. A simple RESTful API for gathering information about a server, in my
mind anyway, was along the lines of:

/views -- returns a list of views
/views/_default -- returns information about a specific view
/views/_default/zones -- returns a list of zones in a particular view
/views/_default//zones/isc.org -- returns detailed information about this
particular zone

and so on. I am not certain if the /config, /stats, or /status information
should be before or after the URIs in this simple example. My gut tells me
after.

--Michael