write_buf_size > buffer_size_reproxy_url

[Eric Lambrecht]

Eric Lambrecht wrote:
> Have anybody ever seen a problem where ClientProxy is reproxying a file 
> and the write_buf_size is larger than what should be allowed by 
> buffer_size_reproxy_url?

...continuing in the long tradition of responding to my own emails... ;-)

I figured this one out. When a ClientProxy gets a REPROXY_URI response, 
it does a 'start_reproxy_uri' call followed by a 'next_request' call. If 
a new BackendHTTP object is immediately assigned to the ClientProxy 
during that 'start_reproxy_uri' call (and not later, due to some event), 
then the 'close' call in 'next_request' clobbers the reference to the 
new backend in the ClientProxy.

It's only obvious, isn't it? ;-)

Here's a (simple, but perhaps not ideal) patch that just tells 
BackendHTTP to not clobber references to other backends.

Eric...
-------------- next part --------------
A non-text attachment was scrubbed...
Name: close.patch
Type: text/x-patch
Size: 420 bytes
Desc: not available
Url : http://lists.danga.com/pipermail/mogilefs/attachments/20051103/67759915/close.bin