[zeromq-dev] REQ-REP with "partitioning" backend

[Julio Viegas]

Hello guys,

   I'm using ZeroMQ since the last year as the fabric for connecting a
system composed of multiple Raspberry PIs and a Linux server.  We are about
to launch the first product to a small group of machines, but our second
customer is a big one, so I'm here asking for help to scale our solution
out... :)

   As a good beginner, I decided to employ REQ(PIs)/REP(Linux) for a start.
My backend(SQLite because I need subsecond replies) is not
multithread-friendly for write, so I realized I needed to partition some of
my tables in order to handle more requests.

   I digged the manual after some example of a topology that fits my
problem, but could not find nothing. I will try to explain it bellow:

   Client 1 -->            --> Backend X
   Client 2 --> Router --> Backend Y
   Client 3 -->            --> Backend Z

   The idea here is Client 1 always talk with Backend X(2 w/ Y, 3 w/ Z and
so on...) but everything is connected through this "intelligent" Router in
the middle. This Router reads a header in the package that provides the
"stick" between a client and a backend. This is very analogue to the
concept of "sticky sessions" between a browser and a application server.
For instance I would love to have something like a zmq_proxy() with some
sort of "stickness" on it.

   Any good soul willing to point me a sample of code(or a similar zmq
solution) to this Router I'm in need?

   Thank you so much, have a nice week!

Rgrds,
JV -- julioviegas.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20140419/a30b1eea/attachment.htm>