[erlang-questions] Why Beam.smp crashes when memory is over?

[Tony Rogvall]

Not yet.

What kind of flavor do you have in mind?

One proposal is to let sender crash when receiver in box is full.

I would like to test a blocking version as well. This may sound like  
utterly
crazy, but we already have the infamous busy_port ;-)
Busy ports do block senders (mostly distribution stuff).

An other flavor is to drop the overflow message. This can however be  
done
by catch (Pid ! Message) if the crash semantics is implemented.

A suggestion from Patrik Winroth is to be able to set thresholds on  
resource limits,
say in percent from max limit. When threshold is reached then a  
warning message
is sent to a logger process.

The max_memory limit currently control number of words! that can be  
allocated for heaps.
This is the sum of sizes of all heaps for the process and it's "sub"- 
processes
(spawned by process that had the limit set in the first place). It is  
implementation specific
if the message queue and/or the messages them selfs are located on the  
process heap.
In the case where messages are none shared and located on a separate  
heap fragments
this should be accounted for.

/Tony




On 10 nov 2009, at 12.18, Chandru wrote:

> 2009/11/9 Tony Rogvall <tony@REDACTED>
>
>> Interesting discussion!
>>
>> I have been working on a resource system for Erlang for nearly two  
>> years
>> now.
>> I have a working (tm) prototype where you can set resource limits  
>> like
>> max_processes/max_ports/max_memory/max_time/max_reductions ...
>>
>
> You don't happen to have a max_message_q_size option do you? Or does
> max_memory implicitly enforce this?
>
> Chandru