[erlang-questions] Delivery of messages

[Joe Armstrong]

Tricky question.

What about time and if the system is very very very slow?

Assume nothing is observed to be broken - you'd have to ask "with no
observations that indicate failure can we assume messages are delivered
within some time T" - but Erlang has no notion of delivering a message
with time T.

The Erlang programming model assumes that in the absence of evidence to
the contrary then messages between a single pair of processes will be
delivered in the order they were transmitted. The underlying
infrastructure is supposed to guantee this - but it would be
impossibly to prove this correct.

Also the destination process must always be available to receive a message
but even if the destination process has not crashed there might be a logic
error in the code so that it would never receive the message (again
which could not be proved)

I use a weaker condition - it you want to be sure that a process
received a message then send a reply back, once you have received the
reply you know
the message was received. You don't have to do this for every message
but whenever you feel appropriate.

/Joe




On Thu, Feb 23, 2012 at 6:49 PM, P.H.Welch <P.H.Welch@REDACTED> wrote:
>
> Hello - I'm new to Erlang but have been working on concurrency stuff
> for 27 years plus.  I'm afraid I want to ask a question asked many
> times before, but under specific conditions that I've not seen nailed
> down.  It's about the guaranteed delivery of messages.
>
> The most authoritative discussion I've found is question 10.10 at:
>
>  http://www.erlang.org/faq/academic.html
>
> The answer quotes Per Hedeland:
>
>  "Delivery is guaranteed if nothing breaks ..."
>
> which would have been enough but for the following discussion!  This
> continues with:
>
>  "It seems this issue of "guaranteed delivery" comes up every now
>   and then, but I've never managed to find out exactly what it is
>   those that are asking for it actually want ..."
>
> Three possible wants are then itemised, none of which are mine.  My
> conditions are:
>
>  If process P sends a message to process Q and:
>
>    * neither process crashes;
>    * Q commits to receive P's message;
>    * the cores/processors/computers on which P and Q are running
>      do not crash or get turned off;
>    * no part of the communication fabric delivering the message
>      crashes or gets turned off;
>
>  is Q guaranteed to receive the message?                       (1)
>
> I want to be sure that, under these conditions (in which nothing is
> breaking), no message is ever lost in the system between P and Q.
>
> A message *might* be lost by a non-crashing communication fabric
> (e.g. if a buffer somewhere became full and, for memory or time
> reasons, it was thought best to discard some traffic silently) or
> by a non-crashing Erlang run-time system (same reason).
>
> Do we have a guarantee that this (or something similar) does not
> happen?
>
> Is the answer to question (1) ... "Yes"?
>
> Thanks,
>
> Peter Welch.
> _______________________________________________
> erlang-questions mailing list
> erlang-questions@REDACTED
> http://erlang.org/mailman/listinfo/erlang-questions