Re: Automatic Reinitialization of connections

On Tue, May 28, 2002 at 04:09:49PM +1200, Malcolm Gibbs wrote:
> RSC card). You notice it automatically retries immediately. The small
> delay between each try is the expect script attempting the connection
> (intentional problem was failed login into RSC). 
> 
> conserver.cf line: 'rschost:|/opt/conserver/bin/rsclogin arg1 arg2::&::'

yep, that explains things.  the problem here is that, as far as
conserver is concerned, everything is ok - for a total of 5 seconds.
then the program exits (which means the console is down), and conserver
respawns it.  the unfortunate part here is that there's no way of
knowing if the program exited because it was having trouble, or of it
exited because it finished normally.  ideally, a "normal exit" would
trigger a respawn and an "abnormal exit" would not - well, maybe.  i've
been thinking that if we look at the exit code of the program and
either auto-respawn that it might be a good compromise.

but i'm not sure about that either.  why?  'cause i would assume that
if something bad happened to a console (say the rsc unit resets), the
program would exit with an error code reflecting that.  and you'd want
conserver to respawn the thing, cause, in theory, the rsc unit would be
back and available "real soon".  with examples like this, you could
argue that the return codes of the programs are irrelevant.  which
brings us back to our original problem - do you respawn these things or
not?

for tcp connections and local serial ports, the current logic is
probably correct (or real close).  for programs, i really don't know
what to do.  i'd think it's safer to error on the side of respawning.
but, maybe we just need another flag that says "don't ever
automatically reconnect on failure" (the -o/-O flags would still take
effect - it would just turn off the initial retry on failure).

if anyone has an opinion on this, let me know (send me private email,
please - i'd rather not spam the list).  if there's a lot of activity,
i'll post a summary.  but, i'm thinking the flag would be the right
thing for the immediate future.  probably only useful for folks who are
using programs as consoles (and, as i said, respawning is probably a
good thing - even for programs) but i can see the need for it not to
respawn too.  (yeah, a mixed console set - some programs, some tcp or
devices - probably won't find this useful as it would turn them all
off, but you could use the -O option.  hmm...)

Bryan