[erlang-questions] Supervisor does not restart Gen Server

[Gunilla Arendt]

Hi,

This is a really tricky one!

My guess is that you start the supervisor from the Erlang shell, right?
Since supervisor only has the start_link alternative, you'll get a link
between the shell process and the supervisor.

Then you call zuzu:do(0), which I would guess calls gen_server:call/2.
The gen_server receives the request, does a division by zero and
subsequently terminates.

The supervisor notices the gen_server has terminated and restarts it.

Unfortunately, since the original gen_server process terminated,
the call to zuzu:do(0) failed. Which means that the shell process
terminates.

And the shell process was linked to the supervisor, which means that
now also the supervisor terminates.

Solution? Calling "catch zuzu:do(0)" should do the trick.

Regards,
Gunilla, Erlang/OTP team

Gurgen Tumanian wrote:
> Hello everybody.
> 
> In my application i have a set of gen_servers orginized in a simple one 
> level supervision tree, with  one superviser above all.  Now the problem 
> that i have , is that once one of my gen_servers is crashed, the 
> supervisor does not restart it, and what is even worse - dies himself.
> I have simplified the problem, creating a simple gen server in which a 
> badarith can be initiated(divide by zero)  and a
> simple supervisor, which should have restarted it in case of crash by 
> one_to_one strategy.
> 
> now when starting the supervisor, and then calling the zuzu:do(0) 
> (causing the badmatch)
> i get the crash of the worker gen_serv, and then the supervisor 
> crashes.Supervior's crash report is
> =CRASH REPORT==== 20-Feb-2007::16:35:12 ===
>   crasher:
>     pid: <0.41.0>
>     registered_name: []
>     error_info: {{badarith,[{zuzu,handle_call,3},
>                              {gen_server,handle_msg,6},
>                              {proc_lib,init_p,5}]},
>                   {gen_server,call,[zuzu_serv,{do,0},35000]}}
>     initial_call: {gen,init_it,
>                       [gen_server,
>                        <0.39.0>,
>                        <0.39.0>,
>                        supervisor,
>                        {self,bubu,[]},
>                        []]}
>     ancestors: [<0.39.0>]
>     messages: []
>     links: []
>     dictionary: []
>     trap_exit: true
>     status: running
>     heap_size: 233
>     stack_size: 21
>     reductions: 227
>   neighbours:
> 
> I get the feeling that the supervisor dies as it pasess the restart 
> limit, but that's a quite surprise for me, as the first start of the 
> worker proc pases normally.
> I will be greatfull for any advice on this issue.
> 
> Regards
> Gurgen Tumanyan, Smart Tech