Issue 86: Patch for requeued requests

[Lieven Govaerts]

Hi,


I was looking at issue 86, and wondered why we do need to keep the
hold_requests queue on a connection.

The purpose seems to be to stop sending requests on a connection when
that connection is pending reset. So if that happens serf moves all
outstanding requests to a separate holding queue.

Instead of this difficult to understand implementation, can we not
just simply keep all pending requests on the connection requests
queue? We can stop writing requests on a connection when its state is
SERF_CONN_CLOSING.

Attached patch shows what I mean, it fixes a memory corruption that I
have seen when forcing the client to renegotiate a ssl session often.

Am I overlooking something?

Lieven

[Justin Erenkrantz]

A quick reply: when I was fixing up stuff in Berlin at the SVN
Hackathon, it struck me that we could likely remove the holdqueue code
and simplify some tricky logic...but, I never got around to it. As
long as the priority requeuing works, we should be good.

Thanks. -- justin