New features, and bugfix for faimond

[Jonas Eriksson]

On Tue, Aug 28, 2007 at 05:14:19PM +0200 Thomas Lange wrote:
> >>>>> On Tue, 28 Aug 2007 16:02:36 +0200, Jonas Eriksson <zqad at hpc2n.umu.se> said:
> 
>     > This signal is just used to abort the read(). Since the die make
>     > script fall through the read, the client is disconnected and the
>     > daemon keeps on running.
> Since I do not want to use the daemon version of your faimond, a die call
> will kill the faimond itself. I like to close the connection, not kill
> the script.

I'm unable to trigger this while running "faimond -d -l -", at
least with the latest set of patches I submitted. All that seems
to happen is that the client gets disconnected.

>     > I must say that I am a bit sceptical to let the faimond-gui lead
>     > the development of faimond. My belief is that most sites will use
>     > faimond without the GUI.
> The faimond was created for the GUI, not for writing log messages into
> a file. Maybe you want to use the faimond also for this purpose, but
> that was not my intension. Maybe we need two seperate scripts:

> faimond-gui (which may include the faimond code of receiving the
> messages via socket) and fai-log-daemon.
> 
> But then not both scripts may run at the same time because of the port
> they listen to.
> 
> I like to release a first version of the faimond-gui soon, this is
> more important for me than logging messages to a file. Those logging
> messages are IMO now interesting, since we have the log files after an
> installation, and I only want to monitor multiple installation via a
> GUI and forget those information if the installations are finished.

I did not know of this intention with faimond. We mainly use
faimond to run "chboot -d" on installed clients and logging, and
believe that we can make these two alternatives co-exist in the
same script, unless faimond should be assimilated in it's whole
by the GUI executable.

So, how should i design the patch? Should the standard behaviour
of faimond be to not fork and not print timestamps? I'm very keen
to get this into the standard release since it's getting tiresome
to port the patches to every new version. To have to supply the
faimond with additional command line options is however an
acceptable solution for us.

I believe that many organizations today (I know three or probably
four without leaving my campus, and about 5 off-campus) is mostly
interested in using the simple log-and-chboot-functions.

>     > Proc:Daemon package is to avoid reinventing the wheel and by
>     > that also having more code with potential bugs. It also makes the
>     > code more readable, IMHO.
> You are right. I agree with you, but I my use case it's not needed to
> run faimond as a "real" daemon.

I guess this depends much on the design of the GUI, and that it
in this case should start an own instance of faimond and read
directly from it's standard out. However, the possibility to
start faimond as a daemon and write to a fifo or a log over nfs
that the gui client can be interesting IMO. Even though it's not
needed for that specific use case today, it opens the gates for
many interesting uses in the future, while keeping us other folks
happy.

Best regards,
Jonas

-------------------------------------
- Jonas Eriksson, zqad at hpc2n.umu.se -
- sysadmin @ hpc2n.umu.se           -
-------------------------------------