cpgx: fix sync

the synced_nodes and sync_from weren't accounting for nodes that had
been removed.  Problems here found by Honza.

Signed-off-by: David Teigland <teigland@redhat.com>

1 files changed, 45 insertions, 22 deletions

diff --git a/cpgx/cpgx.c b/cpgx/cpgx.c
index 12c824e..dd8c12a 100644
--- a/cpgx/cpgx.c
+++ b/cpgx/cpgx.c
@@ -237,10 +237,14 @@ static void _log_config(struct dct_config *c, uint32_t id, int error)
 	for (i = 0; i < c->left_count; i++)
 		off += sprintf(l_buf+off, " %u", c->left[i]);
 
-	if (error)
+	if (error == 1)
 		log_error("%08u conf %u %u %u memb%s join%s left%s",
 			id, c->memb_count, c->join_count, c->left_count,
 			m_buf, j_buf, l_buf);
+	else if (error == 2)
+		log_debug("%08u conf %u %u %u memb%s join%s left%s",
+			id, c->memb_count, c->join_count, c->left_count,
+			m_buf, j_buf, l_buf);
 	else
 		log_history("%08u conf %u %u %u memb%s join%s left%s",
 			id, c->memb_count, c->join_count, c->left_count,
@@ -528,21 +532,6 @@ void update_nodes_list(struct dct_config *c, uint32_t id)
 	uint8_t low = 0;
 	int i, is_memb, is_join, is_left;
 
-	i = 0;
-	memset(&synced_nodes, 0, 8);
-
-	list_for_each_entry(node, &nodes, list) {
-		if (!node->is_member)
-			continue;
-		if (node->needs_sync)
-			continue;
-
-		synced_nodes[i++] = node->nodeid;
-
-		if (!low || node->nodeid < low)
-			low = node->nodeid;
-	}
-
 	list_for_each_entry(node, &nodes, list) {
 
 		is_memb = in_memb(c, node->nodeid);
@@ -589,9 +578,7 @@ void update_nodes_list(struct dct_config *c, uint32_t id)
 
 			node->is_member = 1;
 			node->needs_sync = 1;
-			node->sync_from = low;
 			node->join_eventid = id;
-			memcpy(&node->synced_nodes, &synced_nodes, 8);
 			continue;
 		}
 
@@ -600,6 +587,36 @@ void update_nodes_list(struct dct_config *c, uint32_t id)
 			  node->is_member, node->nodeid);
 	}
 
+	/* figure out who should send syncs and what synced_nodes are */
+
+	i = 0;
+	memset(&synced_nodes, 0, 8);
+
+	list_for_each_entry(node, &nodes, list) {
+		if (!node->is_member)
+			continue;
+		if (node->needs_sync)
+			continue;
+
+		synced_nodes[i++] = node->nodeid;
+
+		if (!low || node->nodeid < low)
+			low = node->nodeid;
+	}
+
+	if (!low)
+		log_error("update_nodes_list no synced nodeid");
+
+	/* old nodes that have joined again */
+
+	list_for_each_entry(node, &nodes, list) {
+		if (node->is_member && node->needs_sync &&
+		    node->join_eventid == id) {
+			node->sync_from = low;
+			memcpy(&node->synced_nodes, &synced_nodes, 8);
+		}
+	}
+
 	/* add new node we've not seen before */
 
 	for (i = 0; i < c->memb_count; i++) {
@@ -610,15 +627,15 @@ void update_nodes_list(struct dct_config *c, uint32_t id)
 		node = add_node(c->memb[i]);
 		node->is_member = 1;
 		node->needs_sync = 1;
-		node->sync_from = low;
 		node->join_eventid = id;
+		node->sync_from = low;
 		memcpy(&node->synced_nodes, &synced_nodes, 8);
 
 		is_join = in_join(c, node->nodeid);
 		if (!is_join)
 			log_error("join list off %d", node->nodeid);
 	}
-
+	
 	log_debug("update_nodes_list");
 	print_nodes_list();
 }
@@ -992,6 +1009,8 @@ void save_message(struct dct_header *hd, int len, int type)
 	memcpy(&se->buf, hd, len);
 
 	list_add_tail(&se->list, &saved_events);
+
+	log_debug("save_message");
 }
 
 void save_confchg(struct dct_config *c, int we_join)
@@ -1013,6 +1032,9 @@ void save_confchg(struct dct_config *c, int we_join)
 	memcpy(&se->buf, c, sizeof(struct dct_config));
 
 	list_add_tail(&se->list, &saved_events);
+
+	log_debug("save_confchg");
+	log_config(c, 2);
 }
 
 static int nodeid_compare(const void *va, const void *vb)
@@ -1095,7 +1117,6 @@ static void confchg_cb(cpg_handle_t handle, const struct cpg_name *group_name,
 		cpg_finalize(our_cpg_handle);
 		client_dead(our_cpg_client);
 		add_history_confchg(&c);
-		update_nodes_list(last_config, last_config_eventid);
 		return;
 	}
 
@@ -1598,8 +1619,10 @@ void loop(void)
 			}
 		}
 
-		if (timeout_sec && (time(NULL) - last_dispatch > timeout_sec))
+		if (timeout_sec && (time(NULL) - last_dispatch > timeout_sec)) {
 			log_error("no cpg dispatch in %d sec", timeout_sec);
+			print_nodes_list();
+		}
 
 		if (got_error) {
 			dump_write();