[erlang-questions] How long a NIF can block the Erlang BEAM?

[Alceste Scalas]

 On Tue, 23 Nov 2010 17:34:07 +0900, Kenji Rikitake 
 <kenji.rikitake@REDACTED> wrote:
> I wonder how long a NIF can block the Erlang BEAM.  To put it another
> way, I want to know the maximum time allowed a NIF can run in the 
> BEAM.
> One millisecond? Or 100 microseconds? Or even shorter?
>
> I am asking this to find out if I can run a fairly large operation of
> initialized a random number table in a NIF.

 AFAIK, there are no hard limits: a NIF could block its Erlang
 VM thread as long as needed --- but long-running NIFs will
 disrupt the scheduling of other Erlang processes, and reduce
 the responsiveness of unrelated applications running in the
 same VM.

 If you need to perform a long operation, and latency is not
 an issue, you could spawn a dedicated NIF thread (using
 enif_thread_create()), and let it send its return value to
 the calling Erlang process (using enif_send()).

 Thus, in your sfmt.erl file, you could do something like this:

     init_gen_rand() ->
         unexported_nif_init_gen_rand(self()),

         receive
             InitGenRandResult ->
                 InitGenRandResult
         end.

 This way, init_gen_rand/0 will still block until the PRNG is
 ready, but its Erlang process will be scheduled away.

 BTW: if you use this strategy, you may also need to perform
 the PRNG initialization within the "on_load" function
 (otherwise, you may need to protect the PRNG table with
 locks...).

 Regards,
-- 
 Alceste Scalas <alceste@REDACTED>