You might like to try this patch. It seems to reduce the number of cman retries
considerably in my testing.
--
patrick
Index: cman-kernel/src/cnxman.c
===================================================================
RCS file: /cvs/cluster/cluster/cman-kernel/src/cnxman.c,v
retrieving revision 1.51
diff -u -p -r1.51 cnxman.c
--- cman-kernel/src/cnxman.c 25 Feb 2005 10:16:36 -0000 1.51
+++ cman-kernel/src/cnxman.c 25 Feb 2005 14:48:44 -0000
@@ -340,10 +340,6 @@ static int cluster_kthread(void *unused)
if (quit_threads)
break;
- if (test_and_clear_bit(ACK_TIMEOUT, &mainloop_flags)) {
- check_for_unacked_nodes();
- }
-
/* Now receive any messages waiting for us */
spin_lock_irq(&active_socket_lock);
list_for_each_safe(socklist, temp, &active_socket_list) {
@@ -368,6 +364,10 @@ static int cluster_kthread(void *unused)
}
spin_unlock_irq(&active_socket_lock);
+ if (test_and_clear_bit(ACK_TIMEOUT, &mainloop_flags)) {
+ check_for_unacked_nodes();
+ }
+
/* Resend any unacked messages */
if (test_and_clear_bit(RESEND_NEEDED, &mainloop_flags)
&& acks_expected) {
