Re: errors when message size exceeds SERVER_BUFSIZE?

I haven't seen my post come back to me from the list, though I do see  
it in the archives on the web.  Is the list not operating properly?   
Hopefully this is getting to someone who has a clue...

Last night I wrote:

> We're using mod_fastcgi with Apache 1.3 on Mac OS X.  All is  
> working fine for small responses, but for larger ones... we find:
>
> [Wed Mar 12 21:22:22 2008] [error] [client 127.0.0.1] FastCGI: comm  
> with server "<path>" aborted: idle timeout (30 sec)
>
> Poking through the source, I see that SERVER_BUFSIZE is set to  
> 8192.  But the fastcgi protocol clearly allows messages to be up to  
> (almost) 64K.  So wouldn't it be a violation of the protocol for  
> mod_fastcgi to fail on messages bigger than 8K?

After some more exploration, I find that it's not this simple.  For  
messages over 16K or so, it works fine.  It also works fine for  
messages under 8K.  Some observations:

Data length	Result
-----------	------
7975		OK
8183		OK
8262		fail
8360		fail
10925		fail
12680		fail
12860		fail
14480		fail
16010		fail
16325		OK
16550		OK
18080		OK
18305		OK

I have my application logging all the fastcgi communications it sends  
or receives, and there is no difference between the "fail" and the  
"OK" cases, apart from the amount of data it's sending (as declared  
in the message header, and verified as actually sending), and the  
data itself.  So this sure looks like it must be a fastcgi bug...   
anyone have any insight on it?

I'm leery of simply increasing SERVER_BUFSIZE, since seems likely to  
just shift the range of sizes for which the failure occurs.

Thanks,
- Joe


___________________________________
fastcgi-developers mailing list
http://fastcgi.com/fastcgi-developers/