[Cluster-devel] cluster/group/daemon app.c

[teigland@sourceware.org <teigland@sourceware.org>]

CVSROOT:	/cvs/cluster
Module name:	cluster
Changes by:	teigland at sourceware.org	2006-10-12 14:25:59

Modified files:
	group/daemon   : app.c 

Log message:
	Handle the case where we're the second node being added to the group
	and the only other member fails.  We need to go ahead and process
	our join.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/group/daemon/app.c.diff?cvsroot=cluster&r1=1.49&r2=1.50

--- cluster/group/daemon/app.c	2006/10/06 16:55:19	1.49
+++ cluster/group/daemon/app.c	2006/10/12 14:25:58	1.50
@@ -1427,7 +1427,7 @@
 {
 	app_t *a = g->app;
 	event_t *ev, *rev;
-	node_t *node;
+	node_t *node, *us;
 	struct save_msg *save;
 	struct nodeid *id, *safe;
 	int rv = 0;
@@ -1505,6 +1505,31 @@
 		log_group(g, "rev %d is for group we're waiting to join",
 			  rev->nodeid);
 
+		/* If the failed node is the only other app member apart
+		   from us in the pending membership list, then we must go
+		   ahead with our own join event, there will be no remote nodes
+		   processing a rev or an ev for this group.  We send a recover
+		   message so other nodes waiting to join after us will purge
+		   their rev on the group. */
+
+		if (a->node_count == 2) {
+			node = find_app_node(a, rev->nodeid);
+			us = find_app_node(a, our_nodeid);
+
+			if (node && us) {
+				log_group(g, "joining group with one other node"
+					  " now dead rev %d", rev->nodeid);
+				a->node_count--;
+				list_del(&node->list);
+				free(node);
+				send_recover(g, rev);
+				del_recovery_set(g, rev, 1);
+				list_del(&rev->list);
+				free_event(rev);
+				return 0;
+			}
+		}
+
 		/* Look for a remote node with stopped of 1, if we find one,
 		   then fall through to the 'else if (event_state_stopping)'
 		   below.  A remote node with stopped of 1 means we've received