Hi all, I've been looking at https://issues.apache.org/jira/browse/TS-1062, where the VC_EVENT_READ_COMPLETE event is not propagated up through the PluginVC.
When the HTTP engine reads the final response chunk, we end up in HttpSM::tunnel_handler_server(), and I think that this is the point where VC_EVENT_READ_COMPLETE needs to be propagated. One way to fix this is to shut down the tunnel VC completely by calling do_io_close() on it. This would make the VC_EVENT_READ_COMPLETE handling roughly similar to VC_EVENT_EOS. I guess that this may prevent some re-use of the origin server connection. But I think that the right way would be to tell the HttpTunnelProducer's vc that the read is complete. AFAICT, VConnection implementations only deliver VC_EVENT_READ_COMPLETE as a side-effect of doing I/O on it, so I'm not really sure how to arrange this. I'd sure appreciate any advice or pointers in the right direction ... cheers, James
