On 30/07/2012, at 07.15, Nathan Duran wrote:
[..snip9
>> since mongrel2 has proxy support, I would simply implement this operation by 
>> first starting the app-specific http server and then *dynamically* changing 
>> the mongrel2 config by adding a proxy for that server. The client would then 
>> connect to the app-specific server via the proxy. Is that possible?  Does it 
>> sound reasonable?
> 
> That does not sound very reasonable at all and may even be a recipe for 
> remote fork bombs. Why would you even want to build something that way? 

Nathan, thanks for replying. I should have been more clear here. The service is 
a kind of management service so the starting/stopping of the app-specific http 
service is actually what it is supposed to do. For other reasons the remote 
fork bomb is not possible in this service.

What I was interested in is if Mongrel2 would support such dynamic 
re-configuration of proxies without needing restart or slowing other operations 
down (the http servers dynamically appear and disappear). 

>> - Finally a questions about security. I see that Mongrel2 easily supports 
>> SSL. But is there any other features that make security easier. Particularly 
>> I'm looking to do authorization at the service level. Each of those Ruby 
>> services need to authorize all requests that come in - whether they come 
>> from the web service or internally. Would all http related information (e.g. 
>> Basic Auth) be relayed from mongrel2 to the ZMQ server processes? How could 
>> a non-mongrel zmq process connect to a service and send the same auth 
>> information (was that question clear)?
> 
> Regardless of the server in question, you're better off implementing your own 
> auth mechanisms if you want anything to actually be secure. I've vended 
> JSON-RPC endpoints that required embedded tickets, and I'm sure you could do 
> something similar with OAuth headers or whatever else you wanted. There is no 
> difference between a Mongrel ØMQ process and a non-Mongrel ØMQ process as far 
> as ØMQ is concerned. I'd strongly recommend working through some of the 
> example code before making any architectural decisions. They're all very 
> short and very simple.

OK, thanks for the input. I definitely will.

Reply via email to