> > The command exits with a non-zero return code, and the application
> > invoking it is required to cope with the error and try a different
> > method of signaling a problem. If it doesn't cope, the message is
lost.
> 
> Goodbye system logs. cron runs stuff like this:
> <cronjob> 2>&1 | mail ...

No argument there. Note comments on smarthost reliability. 

> > Remember the basic assumption was that he *didn't* want to run a
full
> > MTA all the time. Starting with that assumption, you have to give up
> > some of the niceties of a full MTA daemon, like queuing and
automated
> > retries, because there isn't any process there to DO the queuing or
> > retries. If immediate delivery fails, your application has to make
the
> > call on what to do about it -- the MTA isn't there to defer things
for
> > you.
> Indeed. _I_ wouldn't accept the consequences. I want my logs,
especially
> when something went wrong.

Which is why I use a central logging host, and do all the logging
produced by clients to that remote host. If a client tanks, I don't want
to have to dig through the remains; I want the diagnostics somewhere
central and easily located, and better yet, preparsed and ready to go,
suitably edited and processed for analysis. 

I don't log anything on clients locally. If the client is so horked that
it can't send syslog output, it's not going to give me anything useful
to help fix it anyway, and I've done the up-front network engineering to
ensure that UDP packets don't get dropped (I wish syslog-ng would
stabilize...TCP would be good...). Crash logs on the console get
recorded where I have easy facilities to do so (yay, VM spooling
system!), but everything else goes to the logging host. 

> I don't think you want your logwatch report written to syslog, do you?
I
> don't. There are other detailed reports too, no doubt.

See above. I want that stuff in a central place, not scattered about.
All that analysis gets done on the log host, not on the clients.

> As I understand it, SMTP is a store-and-forward proposition, Since
some
> MUAs (and in particular the mail command) cannot handle failure, then
> the MTA _must_ accept the initial submission.

No, it doesn't *have* to. It's *desirable*, but an implementation that
does not queue is completely conformant to the SMTP protocol spec if it
does not do so. The SMTP protocol is connection-oriented like all the
other TCP services -- no store and forward assumption at all. Queueing
is a feature that is implemented by particular applications that provide
the protocol support, but is not required by the spec. Is queuing
useful? Sure. Is it required? No. 

I will agree that *if* you are using mail as a transport for other
semantic data (since Unix never really developed the idea of a real
sender-initiated file transfer capability like NJE, thus used mail/uucp
to substitute), then you want mail to queue, but that's a implementation
detail/decision, not a requirement. 

I think we are looking at two different philosophies of how to do system
management here. I'm from the "collect it all in one place and analyze
it there" school, you're from the more distributed analysis school.
Neither are wrong; they just start from different assumptions. 

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to