Here is a more concrete example, I think may people have encountered the
same problem:

Suppose process A opens a local file and get some file descriptor fd. It
then passes the fd as message over the network. Now suppose process B
gets the message and tries to read from fd, something bad might happen
unless B is actually A (or forked from A etc.):

- It can encounter the EBADF exception since there is no such file
descriptor opened on by this process
- Even worse, if it happens to have a fd open locally with the same fd
number, it may succeed in reading but fail later for other unexpected
reasons.

Now suppose we're allowed to prefix the fd with some tag information to
ensure safety, so that when a process receives this fd, its reading
function can at least check:

- whether it was generated by myself?
- whether it was generated by another process on the same machine ?
(suppose you also pack the file information into the tag, then there
might be something this process can do)
- or it was generated on other machines (i.e. nothing we can do here,
maybe just passing the fd on and on)

Even if we are forced to open a fd no matter it was generated on the
local machine or not, the raised exception can at least carries much
clearer message than EBADF or others, such as "Can't read: Fd was
generated on machine XXX".

Well we may still encounter exception when a local fd was previously
open but now closed, but this is a problem even in purely local
applications. So it's not a problem newly introduced by the "passing fd
as message" scheme, hence not my concern.

The question is what the tag should look like?

Thanks.

code17 wrote:

> I'm using OCaml to implement a library for a network application where
> many distributed processes running and communicating. A process can
> generate and send messages to others, or pass messages it receives
> *without* any modification.
>
> The only requirement is that when a process gets a message, it must be
> able to identify whether this message was previously generated by itself
> (and circulate back), or by another process but running on the same
> machine, or by processes on other machines.
>
> For that, we are allowed to prefix message strings with any kind of tag
> when it's generated. The problem is what the tag should look like? Even
> simpler, we can assume all the processes start and end at the same time,
> so there is no problems about pid/IP reuse.
>
> My naive attempt is using hostname + pid as the tag, but two machines
> can named themselves the same but still communicate well if they choose
> to always connect through explicit IP, can't they?
>
> If I instead use ip + pid, how can I detect the ip of a running process
> without forcing users to always provide one explicitly?
> Unix.gethostbyname(Unix.gethostname()) only returns 127.0.0.1, and what
> if a machine has several IPs?
>
> The problem turns out be what is unique for a host machine in a
> networked application? It seems simple but I don't really know what's
> the common idiom. Any suggestion is welcome. Thanks in advance.
>
> --
> code17
>

•  2
  New Members

> My naive attempt is using hostname + pid as the tag, but
> two machines can named themselves the same but still communicate
> well if they choose to always connect through explicit IP,
> can't they?

Hi,

You are looking for some form of UUID:

http://en.wikipedia.org/wiki/UUID
http://erratique.ch/software/uuidm

Hope this helps,
Dario Teixeira

•  2
  New Members

In addition to the other reply, there is an RFC 4122-compliant UUID
generator libary for OCaml which you may find useful:

http://alan.petitepomme.net/cwn/2008.06.17.html#1

Rich.

--
Richard Jones
Red Hat

•  2
  New Members