Le 14/04/2014 21:39, Markus Bergkvist a écrit :
What about using websockets?
I don't know them. I will check, thanks.
Out of curiosity, how do you use the CURVE security mechanism from the
client?
First approach, a three layers protocol with ZMQ_STREAM:
http://lists.zeromq.org/pipermail/zeromq-dev/2014-February/025168.html
It works (the last working version is not published - I will do it
someday when I have time).
Finally, I have opted to put CURVE in the application layer, like libcurve.
In fact, I don't use CURVE, but a security protocol of mine, with
registered clients who know already the server public key, exchanged
during registration via https.
Cheers,
Laurent
/Markus
Den 14 apr 2014 12:37 skrev "Laurent Alebarde" <[email protected]
<mailto:[email protected]>>:
Hi,
If my understanding is correct, Mongrel2 strength reveals in
stateless architectures, where many workers are transparently
connected to the Mongrel2 server which dispatch the jobs with a
0MQ push/pull.
In my use case, I make use of the CURVE security mechanism, so I
need to have a sticky connection between clients and workers
because of the security protocol and keys exchanges. In other
words, when a new client comes, I have to assign it a worker, and
keep sending messages from this client always to this assigned
worker, until the connection is finished.
Can I do that with Mongrel2 please ? Or am I obliged to send all
the messages to an intermediate broker that perform the
client/worker pairing ?
As an alternative, I could put all the connections states,
including keys, in a database shared by all the workers, in order
to stay stateless, but I would have to redesign everything and I
think performances would be worse. So I exclude this solution.
Cheers,
Laurent
