When firing up a new child process, wipe clean the set of worker_score records

we're going to use as they could contain leftovers from a previous child.
This solves a problem where a non-graceful restart followed by a graceful
restart would leave previous generation children stuck waiting for non-existant
worker threads to terminate.

git-svn-id: https://svn.apache.org/repos/asf/httpd/httpd/trunk@93548 13f79535-47bb-0310-9956-ffa450edef68

diff --git a/server/mpm/mpmt_os2/mpmt_os2_child.c b/server/mpm/mpmt_os2/mpmt_os2_child.c
index 47e01564c0d853585bd90e8c402ed3223759b278..60b747300cd3160b4d773ea485b53edcb8bc7818 100644 (file)
--- a/server/mpm/mpmt_os2/mpmt_os2_child.c
+++ b/server/mpm/mpmt_os2/mpmt_os2_child.c
@@ -196,6 +196,7 @@ void ap_mpm_child_main(apr_pool_t *pconf)
     }
 
     ap_my_generation = ap_scoreboard_image->parent[child_slot].generation;
+    memset(ap_scoreboard_image->servers[child_slot], 0, sizeof(worker_score) * HARD_THREAD_LIMIT);
 
     /* Set up an OS/2 queue for passing connections & termination requests
      * to worker threads