Send duplicate multi meet packet only for node which supports it (#2840)

This prevents crashes on the older nodes in mixed clusters where some
nodes are running 8.0 or older. Mixed clusters often exist temporarily
during rolling upgrades.

Fixes: #2341 

Signed-off-by: Harkrishn Patro <[email protected]>

Author: hpatro

src/cluster_legacy.c

@@ -1145,7 +1145,8 @@ void clusterUpdateMyselfFlags(void) {
     myself->flags |= nofailover;
     myself->flags |= CLUSTER_NODE_EXTENSIONS_SUPPORTED |
                      CLUSTER_NODE_LIGHT_HDR_PUBLISH_SUPPORTED |
-                     CLUSTER_NODE_LIGHT_HDR_MODULE_SUPPORTED;
+                     CLUSTER_NODE_LIGHT_HDR_MODULE_SUPPORTED |
+                     CLUSTER_NODE_MULTI_MEET_SUPPORTED;
     if (myself->flags != oldflags) {
         clusterDoBeforeSleep(CLUSTER_TODO_SAVE_CONFIG | CLUSTER_TODO_UPDATE_STATE);
     }
@@ -3675,6 +3676,13 @@ int clusterProcessPacket(clusterLink *link) {
         } else {
             sender->flags &= ~CLUSTER_NODE_LIGHT_HDR_MODULE_SUPPORTED;
         }
+
+        /* Check if the node can handle multi meet packet. */
+        if (flags & CLUSTER_NODE_MULTI_MEET_SUPPORTED) {
+            sender->flags |= CLUSTER_NODE_MULTI_MEET_SUPPORTED;
+        } else {
+            sender->flags &= ~CLUSTER_NODE_MULTI_MEET_SUPPORTED;
+        }
     }
 
     /* Update the last time we saw any data from this node. We
@@ -5727,7 +5735,8 @@ static int clusterNodeCronHandleReconnect(clusterNode *node, mstime_t now, long
     }
     if (nodeInNormalState(node) && node->link != NULL && node->inbound_link == NULL &&
         now - node->inbound_link_freed_time > getHandshakeTimeout() &&
-        now - node->meet_sent > getHandshakeTimeout()) {
+        now - node->meet_sent > getHandshakeTimeout() &&
+        nodeSupportsMultiMeet(node)) {
         /* Node has an outbound link, but no inbound link for more than the handshake timeout.
          * This probably means this node does not know us yet, whereas we know it.
          * So we send it a MEET packet to do a handshake with it and correct the inconsistent cluster view.

src/cluster_legacy.h

@@ -49,19 +49,22 @@ typedef struct clusterLink {
 #define linkSupportsExtension(link) ((link)->flags & CLUSTER_LINK_EXTENSIONS_SUPPORTED)
 
 /* Cluster node flags and macros. */
-#define CLUSTER_NODE_PRIMARY (1 << 0)                      /* The node is a primary */
-#define CLUSTER_NODE_REPLICA (1 << 1)                      /* The node is a replica */
-#define CLUSTER_NODE_PFAIL (1 << 2)                        /* Failure? Need acknowledge */
-#define CLUSTER_NODE_FAIL (1 << 3)                         /* The node is believed to be malfunctioning */
-#define CLUSTER_NODE_MYSELF (1 << 4)                       /* This node is myself */
-#define CLUSTER_NODE_HANDSHAKE (1 << 5)                    /* We have still to exchange the first ping */
-#define CLUSTER_NODE_NOADDR (1 << 6)                       /* We don't know the address of this node */
-#define CLUSTER_NODE_MEET (1 << 7)                         /* Send a MEET message to this node */
-#define CLUSTER_NODE_MIGRATE_TO (1 << 8)                   /* Primary eligible for replica migration. */
-#define CLUSTER_NODE_NOFAILOVER (1 << 9)                   /* Replica will not try to failover. */
-#define CLUSTER_NODE_EXTENSIONS_SUPPORTED (1 << 10)        /* This node supports extensions. */
-#define CLUSTER_NODE_LIGHT_HDR_PUBLISH_SUPPORTED (1 << 11) /* This node supports light message header for publish type. */
-#define CLUSTER_NODE_LIGHT_HDR_MODULE_SUPPORTED (1 << 12)  /* This node supports light message header for module type. */
+#define CLUSTER_NODE_PRIMARY (1 << 0)                                             /* The node is a primary */
+#define CLUSTER_NODE_REPLICA (1 << 1)                                             /* The node is a replica */
+#define CLUSTER_NODE_PFAIL (1 << 2)                                               /* Failure? Need acknowledge */
+#define CLUSTER_NODE_FAIL (1 << 3)                                                /* The node is believed to be malfunctioning */
+#define CLUSTER_NODE_MYSELF (1 << 4)                                              /* This node is myself */
+#define CLUSTER_NODE_HANDSHAKE (1 << 5)                                           /* We have still to exchange the first ping */
+#define CLUSTER_NODE_NOADDR (1 << 6)                                              /* We don't know the address of this node */
+#define CLUSTER_NODE_MEET (1 << 7)                                                /* Send a MEET message to this node */
+#define CLUSTER_NODE_MIGRATE_TO (1 << 8)                                          /* Primary eligible for replica migration. */
+#define CLUSTER_NODE_NOFAILOVER (1 << 9)                                          /* Replica will not try to failover. */
+#define CLUSTER_NODE_EXTENSIONS_SUPPORTED (1 << 10)                               /* This node supports extensions. */
+#define CLUSTER_NODE_LIGHT_HDR_PUBLISH_SUPPORTED (1 << 11)                        /* This node supports light message header for publish type. */
+#define CLUSTER_NODE_LIGHT_HDR_MODULE_SUPPORTED (1 << 12)                         /* This node supports light message header for module type. */
+#define CLUSTER_NODE_MULTI_MEET_SUPPORTED CLUSTER_NODE_LIGHT_HDR_MODULE_SUPPORTED /* This node handles multi meet packet.                             \
+                                                                                     Light hdr for module and multi meet were both introduced in 8.1, \
+                                                                                     so we could reduce the same flag value. */
 #define CLUSTER_NODE_NULL_NAME                                                                                         \
     "\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000" \
     "\000\000\000\000\000\000\000\000\000\000\000\000"
@@ -75,6 +78,7 @@ typedef struct clusterLink {
 #define nodeFailed(n) ((n)->flags & CLUSTER_NODE_FAIL)
 #define nodeCantFailover(n) ((n)->flags & CLUSTER_NODE_NOFAILOVER)
 #define nodeSupportsExtensions(n) ((n)->flags & CLUSTER_NODE_EXTENSIONS_SUPPORTED)
+#define nodeSupportsMultiMeet(n) ((n)->flags & CLUSTER_NODE_MULTI_MEET_SUPPORTED)
 #define nodeInNormalState(n) (!((n)->flags & (CLUSTER_NODE_HANDSHAKE | CLUSTER_NODE_MEET | CLUSTER_NODE_PFAIL | CLUSTER_NODE_FAIL)))
 
 /* Cluster messages header */