[erlang-questions] : A whole bunch of implementation details

Raimo Niskanen raimo+erlang-questions@REDACTED
Mon Sep 8 12:31:41 CEST 2008


On Thu, Sep 04, 2008 at 10:20:45PM -0500, Dennis Byrne wrote:
> Thank you Richard, Mikael and Raimo.  Here's another question I got as a 
> follow up.
> 
> Can pids be reused, even though the documentation refers to them as 
> unique?  Any info on the underlying implementation of how a pid is 
> uniquely generated across a cluster?

The unique part is a 28 bit field per node. If you spawn processes
in a tight loop you can wrap the 28 bits rather fast, but for
any decent application it will take quite a while.

In a cluster, a particular node name and incarnation identifies
the node, and a remote pid is identified by the node (+incarnation)
and the node's local pid id. A connection between nodes may
go down and come up, and a remote pid still remain valid.

> 
> If they are not unique, what precautions should application developers use 
> to guard against this?

* Short enough timeouts for the process ids not to wrap.
* Short enough timeouts for a remote node node not restart unnoticed.
* Process links and monitors.
* Node monitors.

> ____________________________________________
> Dennis Byrne
> ThoughtWorks - Chicago
> 312-505-7965
> 
> 
> 
> 
> "Richard A. O'Keefe" <ok@REDACTED> 
> 09/04/2008 12:41 AM
> 
> To
> Dennis Byrne <dbyrne@REDACTED>
> cc
> <erlang-questions@REDACTED>
> Subject
> Re: [erlang-questions] A whole bunch of implementation details
> 
> 
> 
> 
> 
> 
> 
> On 4 Sep 2008, at 2:37 pm, Dennis Byrne wrote:
> > How does stack memory grow?  Am told Prolog does this dynamically 
> > and I was wondering if Erlang did the same.
> 
> Some Prologs (DEC-10 Prolog, Quintus Prolog, SICStus Prolog, ...) do 
> this,
> some Prologs (C Prolog, SWI Prolog) don't.
> The Burroughs B6700 used to do this for Algol, BASIC, COBOL, DMAlgol,
> ESPOL, FORTRAN, Pascal, and PL/I, back in the late 60s.
> 
> "Modern", less capable, systems like say Windows or Unix, do not do this
> because they don't know where the pointers are.  The B6700, Prolog 
> systems,
> and Erlang systems, tag each word of memory to say what it is, so you 
> can
> infallibly find the pointers.  With that, moving a stack is tedious 
> rather
> than difficult.
> 
> >  Does heap memory contract?
> 
> It can.
> >
> > Exactly what is the tipping point between what gets passed by value 
> > and what gets passed by reference?
> 
> There is no pass by reference.  All binaries are passed by value. 
> Some binary values are
> represented by pointers.  Some of those pointers point to shared 
> blocks, some don't.  But
> it is still pass by value.  Pass by reference is, by definition, 
> passing the address of a
> VARIABLE ('var x: t' in Pascal, 't& x' in C++).
> 
> Notionally when you send anything in a message, it is copied, 
> otherwise it isn't.
> Some Erlang systems have had a shared or hybrid heap, in which the 
> logical copy is
> a no-op.
> 
> >  The large binary trick is common knowledge, any other corner cases?
> > How does the VM differentiate between a node that is temporarily 
> > unresponsive and one that is down?  In other words, at what point 
> > does a node give up on a peer?
> 
> There is no infallible way to tell.  Mistakes will always be made.
> The net_kernel module sets up a "ticker" process that periodically
> sends a message to every node it's in communication with; if it
> doesn't hear from a node after a while it assumes it's down.
> If a node is shut down in an orderly way it will tell is peers
> that it is going down.
> 
> 
> 

> _______________________________________________
> erlang-questions mailing list
> erlang-questions@REDACTED
> http://www.erlang.org/mailman/listinfo/erlang-questions

-- 

/ Raimo Niskanen, Erlang/OTP, Ericsson AB



More information about the erlang-questions mailing list