https://issues.apache.org/bugzilla/show_bug.cgi?id=56119
--- Comment #31 from Rainer Jung <[email protected]> --- The typical explanation for a Cneonction header is here: http://stackoverflow.com/questions/4798461/cneonction-and-nncoection-http-headers They say it happens if the web server is fronted with a load balancer that does its own connection handling but doesn't want to do to much protocol adjustment work. By permuting some characters in the name of the connection header the client no longer reacts on the close message of the origin server but the packet sizes and probably TCP checksums do not change. So the implementation on the packet level is cheap. The LB reacts on the connection close itself but has decoupled connection handling between client and LB from the connections between LB and origin server. Now if that is true, it might well be, that the LB itself doesn't shut down the connections cleanly once it decides itself to close a connection from the client. That would fit into sebb's observation. Asa result it should be OK to ignore those misspelled headers, because they are misspelled intentionally to get them ignored. And in this situation the next hop might infact expose a somewhat strange connection close behavior itself. -- You are receiving this mail because: You are the assignee for the bug.
