[erlang-questions] beginner: passing messages between supervisor trees

[Ulf Wiger (TN/EAB)]

Pinku Surana skrev:
> I'm still new to the idea of supervisors in Erlang. Assume I have a 
> client C managed by supervisor MC, and server S managed by supervisor 
> MS. When C sends a message to S, does the message get routed through MC 
> and MS first? Or do C and S communicate directly? If so, where in the 
> code are they connected together?

The supervisor has nothing to do with messages going between
C and S. The only job of the supervisor is to monitor its
children, and perhaps restart them if they die.

> If C sends a message to S and S fails, will C's message be saved 
> somewhere so it can try again after restarting S? Is this done 
> automatically, or manually by one of the supervisors?

No.


> If S fails repeatedly, then I want to replace it with S2 (a simpler 
> task). How can C's messages be routed to S2 transparently? I don't 
 > want C to be aware of the failures.

If you want this sort of behaviour, you need some kind of buffering
process between C and S.

The supervisors in Erlang offer no facility for replacing S with
S2 after repeating failures. The main reason is that the arguments
used for (re-)starting S are the same each time, and S is given no
information about why it is restarting, or how many times it's
restarted.

I did make an extended supervisor behaviour once (haven't mentioned
it since end of 2003, so if I'm harping on about it, it's at least
at a very low intensity.) It never caught on, so I guess most people
are perfectly happy with the existing supervisor module. (:

http://www.erlang.org/pipermail/erlang-questions/2003-November/010763.html

BR,
Ulf W