thrift-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Brian Hammond <>
Subject Re: Python server over HTTP, HTTPS -- How?
Date Fri, 03 Apr 2009 17:00:06 GMT
On Apr 3, 2009, at 2:30 AM, Garrett Smith wrote:

> Unless your connections are long running -- to the point that you
> *really* need to support thousands of concurrent connections (this is
> very rarely the case with HTTP as you soon become CPU bound), I'd  
> stick
> with threads -- async can be a serious pain, esp if it starts leaking
> into your app (coroutines, Twisted deferreds, etc.)

Thanks Garrett.  That sounds like good advice.  Perhaps I will take  
this advice until I learn that I really do have very special needs for  
my service.  I'm prematurely optimizing I suppose (in hopes of getting  
it right the first time).

>> Auth in Thrift would be wonderful but I wonder if that's feature
>> creep?
> Yes, definite is. I've been looking at AMQP and their authentication
> scheme, using SASL, is quite simple and still useful. This could serve
> possibly as a model for Thrift. Maybe this has been hashed out  
> before and
> duly rejected though.

I'm not familiar so I looked it up; I suppose you mean this?

There's a lot of back-and-forth chatter there to setup the  
connection.  This kind of thing has always struck me as "too  
much" (sure, the same sort of thing happens in setting up a SSL  
connection) for simply making one-off requests like a RPC.  But then  
again, not everyone is making one-off RPCs I suppose so the connection  
overhead is justified in various cases.

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message