https://issues.apache.org/bugzilla/show_bug.cgi?id=56703
--- Comment #6 from Martin Knoblauch <kn...@knobisoft.de> --- Hi Rainer, good you double checked. I applied my previous fix to a productive system and found that while it reduced the growth by 80-90 %, there was still something missing. So, I found that [in my case] the remaining cases were happening in "ajp_connection_tcp_get_message" when the calls to "jk_tcp_socket_recvfull" were returning JK_SOCKET_EOF. So adding the "connected--" in those places cured the problem for good. I then looked at the remaining case when abort_endpoint was called with "shutdown == JK_FALSE". That is in "ajp_connection_tcp_send_message" when "jk_tcp_socket_sendfull" returns an error. So in my opinion "connection--" is warranted here also. But that would mean that commit r1334399 was the wrong fix to the going negative problem. Hmm... Now looking at r1618000, you have decoupled the checks for "valid socket" and "perform shutdown" in abort_endpoint and decrement the connection counter regardless of the shutdown setting. That will definitely fix the growth problem. But will it also fix the going negative thing? Looking at pre r1334399, a superfluous decrement could happen when abort_endpoint was called with an "invalidated" . Do you this could happen? Anyway, I will try to test on my high volume setup. Martin -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
