slackware-current/source/n/nn/NNTP

			     NNTP SUPPORT
			     ------------

This file describes the NNTP support available in nn release 6.5.  The
NNTP support was implemented by Rene' Seindal, seindal@diku.dk.


			    PREREQUISITES
			    -------------

First of all, you need read-access to an NNTP-server, and if you want
to post, the server must allow that.

If you have news on one of your systems, and want to run an NNTP
server on that system to feed other local systems, you need to get and
install the nntp-1.5 distribution with at least patches 1-3 (I think
patch 8 is the latest).  It is available from several ftp-sites in the
USA.  It is also available on freja.diku.dk (ip 129.142.96.1).

However, just to run nn on you local system with or without NNTP, you
don't need anything besides the nn 6.5 distribution!!

The necessary modules to access a remote NNTP server is an integrated
part of nn, so if you specify to use NNTP, the necessary code is
automatically included.

			     HOW IT WORKS
			     ------------

NNTP is supported both in nn and nnmaster.  When NNTP is used, the
database with the header information used by nn is still maintained on
the local system (because NNTP does not know about the nn database
(yet?)).

When the master is set up to use NNTP, it will connect to the NNTP-
server in each iteration of the collection (the interval set with -r),
get a copy of the active file, and incorporate the new articles into the
database.  To do this, the master will temporarily transfer one article
at a time from the NNTP-server to the local system.

When the articles are read with nn, it will use the local database to
present the menus, and fetch the articles from the NNTP-server as they
are requested by the user.  It will connect to the NNTP server the first
time it is necessary to fetch an article.

Neither nnmaster, nor nn will use NNTP if they run on the NNTP-server
itself (they will directly access the news files).

Both nn and nnmaster access the server in reading mode.  The master and
all client MUST use the same server at all times, since the local
database contains article numbers, that are only unique for each
NNTP-server.


			 SHARING THE DATABASE
			 --------------------

You must also decide whether you want to share the database between your
local news clients, and how you are going to do it.

The database will take up some disk space, normally about 1Mb per 10.000
articles.  There are several ways to manage this space.

This simplest solution, is to let each client run it own master, i.e.,
have its own database.  This means, of course, no sharing.

Alternatively, one host can run the master, and distribute the database
to the others via e.g., rdist.  This doesn't save disk space, but saves
load on the NNTP-server.

Last, the database can be shared with NFS/RFS (see the description of
NETWORK_DATABASE in the config.h file).

The possibility of making a `nndb-server' stands open.  It could be
realized either as a separate server, running under inetd, or it could
be incorporated into nntpd.  It has not been implemented, but might be
part of a future release (any volunteers?).


			    CONFIGURATION
			    -------------

To use NNTP in nn, you must edit the relevant parts of config.h:

NNTP
	You enable the use of NNTP by defining the macro NNTP.

NNTP_SERVER
	Both the master and the clients will look up their NNTP-server
	in the file given by the macro NNTP_SERVER.  If the name is not
	an absolute path name, it is taken to be relative to
	LIB_DIRECTORY.

	The format of the file is compatible with the one used in
	clientlib.c in the nntp-1.5 distribution, i.e., the first
	non-blank line, not starting with '#' is taken to be the name of
	the NNTP-server.  This file MUST be present, and must contain a
	valid host name.

NEWS_LIB_DIRECTORY & INEWS
	If either is defined, they specify the destination of the
	mini-inews program when installed below with INEWS being used
	if both are defined.  If neither is defined, it will be
	installed in /usr/lib/news/inews.


				TUNING
				------

Both the server and each client maintains a cache of recently accessed
articles, to minimize communication with the server (mainly to avoid
fetching large digests continuously).  The master needs the cache when
it splits digests, and the clients need it, because nn has a tendency to
reopen the articles several times.

The master's cache is kept in LIB_DIRECTORY, and each client's cache are
kept in the users .nn directory.  The constant NNTPCACHE (defined in
nntp.c but can be redefined in config.h) defines the size of the cache,
whose optimal size depends on the amount of news kept on line on the
NNTP-server.  Values of 5-10 gives reasonable results.  The effect is
most striking when reading digested news.

The location and size of the cache can also be changed on a per-user
basis via the related nntp- variables (see nn.1).


			     INSTALLATION
			     ------------

Making and installing nn using NNTP does not differ from a non-NNTP nn
installation, except for the differences in the configuration and the
need to specify the NNTP server in the NNTP_SERVER file.

Notice however, that the NNTP_SERVER file must be properly initialized
before doing the 'make initdb'.

If something goes wrong in the initialization of the database, you will
have to run 'nnmaster -I' again by hand.


			    ERROR HANDLING
			    --------------

The handling of errors have been improved since the initial release.

The master will handle most errors by closing the connection, and
returning to the main loop.  All errors in the master are logged, with
a code of `N,' so they can be inspected with the `n' command in
nnadmin's Log menu.

A few errors are considere fatal. If any of these occur operation will
be discontinued.  These errors are such as failure to find the NNTP
server, failure to find the NNTP service, and responses from the NNTP
server in the 500 range (ill-formed requests, access denied, ...)

NNTP server timeouts are handled specially.  If the NNTP server times
out, both nn and the master will attempt to restart it (by connecting
again).  This shouldn't happen in the master (which won't leave sockets
idle for that long), but it can easily happen in nn, if it is left
suspended for too long.  If the server responds with code 400 (Service
discontinued), a reconnect is also tried.


			       PROBLEMS
			       --------

I am not certain what should happen if the server sends back responses
in the 1xx range.  I do not know whether a NNTP server is allowed to
return one of these responses on its own initiative.  If it is, nn
should probably ignore (or display) the messages.  Currently, nothing is
done to treat these responses in any way.

I have seen a strange thing happen to the master, which I have not been
able to reproduce.  The master ran on a Sun-4 running SunOS 4.0, and the
NNTP server was a VAX 785 running MORE/bsd.  The NNTP software was
version 1.5.3.  The master was stuck in a read from the NNTP server.  A
netstat on the Sun show an established connection to nntpd on the Vax,
but a netstat on the Vasx did not show any NNTP connections.  There was
no nntpd running, and no messages on the console indicating any
failures.

[ It is now known that this problem is related to the socket not
  having the KEEP ALIVE flag set, but I have not got the necessary
  patches to fix it,   ++Kim ]


		      SPONTANEOUS NNTP ERROR 502
		      --------------------------

Sometimes nn or nnmaster may stop with the following message:

	NNTP 502 You only have permission to transfer, sorry. 

This particular case is probably the result of the NNTP server trying to
turn your IP address into a fully qualified domain name (FQDN) so it can
look you up in its access file.

The NNTP server probably uses the domain name server (DNS) to map IP
addresses into FQDNs.  If the local DNS doesn't already know the answer, it
has to go out over the network to find it.  This can take a few seconds, and
the library routine that does all this for the NNTP server might time out
before the answer gets back to it.  If this happens, the NNTP server doesn't
know your FQDN, so it gives you the default access specified in the server's
nntp_access file, which is usually "xfer" (article transfer only).

In the time it takes for you to run nn again the DNS usuallu has its answer
back, so things usually work the second time.  

One way to work around this problem is to specify the IP address of the
client in the nntp server's access file; then it is not necessary to lookup
the FQDN.

Thanks to Tim Ramsey and Nick Sayer for this information.


		      DEBUGGING NNTP CONNECTIONS
		      --------------------------

If you want to debug the nntp connection, you can run the nnmaster
with the option -D2 (or -D3 which also turns on the normal -D verbose
output).  In the nn client, you can turn on the nntp-debug variable in
the init file.

The debug output from nnmaster will be placed in $TMP/nnmaster.log
while the output in the client will appear on the message line.


		POSSIBLE EXTENTIONS TO THE NNTP SERVER
		--------------------------------------

The new expire method used in release 6.5 is very efficient on local
systems, because it will just read the spool directories to get a list
of available articles in each group.

However, with nntp, the only way I know of to get a list of available
articles in a group with nntp is the XHDR request.  However, this will
open every article in the group to extract the desired field, but the
only thing I am interested in is the article number itself.

So I suggest to add a LISTGROUP request to the NNTP server to return
the equivalent of
	ls $GROUPDIR | sed -n '/^[0-9][0-9]*$/p'
(in any order - nnmaster will sort the list itself).

Currently nnmaster will test whether this request works before using
the XHDR request, so no changes to nnmaster will be required to take
advantage of such a fix.


Another possible performance increase would be if there was a request
to get the current modification time of the ACTIVE file.  This is the
check nnmaster will do to see if there might be work to do on a local
system, but with NNTP it has to read the active file from the server
and compare it to a local copy to determine whether there is work to
do.  A simple ACTIVESTAT request returning the active file's age and
size would fix this.

Currently nnmaster is not prepared to use such a request, but it would
be easy to add.


			     ALTERNATIVES
			     ------------

Alternative implementations can be conceived, especially in the master.
The master normally collects articles by rereading the active file,
looking for changed article numbers.  For each group with new articles,
it reads the new articles and adds them to the database.  This scheme
has been kept in the NNTP-based master, to keep the changes at a
minimum.

An alternative solution could be to use NEWNEWS to get a list of new
articles since last collect, and fetch each article in sequence.  The
newsgroups and article numbers within each group could then be found in
the Xref: field.  This would probably improve efficiency, since the
master would then generate fewer failing requests (for non-existent
articles), and it would only read cross-posted articles once.  This
solution would, however, require some surgery on the current structure
of the masters main loop.