Bound the number of nodes in gossip section

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

Author: hpatro

src/cluster_legacy.c

@@ -4600,13 +4600,26 @@ void clusterSendPing(clusterLink *link, int type) {
      * Since we have non-voting replicas that lower the probability of an entry
      * to feature our node, we set the number of entries per packet as
      * 10% of the total nodes we have. */
-    wanted = floor(dictSize(server.cluster->nodes) / 10);
-    if (wanted < 3) wanted = 3;
-    if (wanted > freshnodes) wanted = freshnodes;
+    int overall = server.cluster_ping_message_gossip_max_count;
+    if (!overall) {
+        overall = floor(dictSize(server.cluster->nodes) / 10);
+        if (overall < 3) overall = 3;
+    }
 
-    /* Include all the nodes in PFAIL state, so that failure reports are
-     * faster to propagate to go from PFAIL to FAIL state. */
+    /* Prioritize pfail nodes over other nodes.
+     * Healthy nodes can communicate through direct ping/pong if required and failed node
+     * information would be broadcasted. */
     int pfail_wanted = server.cluster->stats_pfail_nodes;
+    if (pfail_wanted >= overall) {
+        pfail_wanted = overall - 1;
+        wanted = 1;
+    } else {
+        wanted = overall - pfail_wanted;
+    }
+
+    if (wanted > freshnodes) {
+        wanted = freshnodes;
+    }
 
     /* Compute the maximum estlen to allocate our buffer. We'll fix the estlen
      * later according to the number of gossip sections we really were able

src/config.c

@@ -3357,6 +3357,8 @@ standardConfig static_configs[] = {
     createUIntConfig("socket-mark-id", NULL, IMMUTABLE_CONFIG, 0, UINT_MAX, server.socket_mark_id, 0, INTEGER_CONFIG, NULL, NULL),
     createUIntConfig("max-new-connections-per-cycle", NULL, MODIFIABLE_CONFIG, 1, 1000, server.max_new_conns_per_cycle, 10, INTEGER_CONFIG, NULL, NULL),
     createUIntConfig("max-new-tls-connections-per-cycle", NULL, MODIFIABLE_CONFIG, 1, 1000, server.max_new_tls_conns_per_cycle, 1, INTEGER_CONFIG, NULL, NULL),
+    createUIntConfig("cluster-ping-message-gossip-max-count", NULL, MODIFIABLE_CONFIG, 0, UINT_MAX, server.cluster_ping_message_gossip_max_count, 0, INTEGER_CONFIG, NULL, NULL),
+
 #ifdef LOG_REQ_RES
     createUIntConfig("client-default-resp", NULL, IMMUTABLE_CONFIG | HIDDEN_CONFIG, 2, 3, server.client_default_resp, 2, INTEGER_CONFIG, NULL, NULL),
 #endif

src/server.h

@@ -2182,6 +2182,7 @@ struct valkeyServer {
     int cluster_port;                                      /* Set the cluster port for a node. */
     mstime_t cluster_node_timeout;                         /* Cluster node timeout. */
     mstime_t cluster_ping_interval;                        /* A debug configuration for setting how often cluster nodes send ping messages. */
+    int cluster_ping_message_gossip_max_count;             /* A configuration for setting how many peer node information to be gossiped in ping/pong messages. */
     char *cluster_configfile;                              /* Cluster auto-generated config file name. */
     struct clusterState *cluster;                          /* State of the cluster */
     int cluster_migration_barrier;                         /* Cluster replicas migration barrier. */

valkey.conf

@@ -2004,6 +2004,12 @@ aof-timestamp-enabled no
 # In order to setup your cluster make sure to read the documentation
 # available at https://valkey.io web site.
 
+# Cluster nodes communicate with each other via PING/PONG message type. Each message carries a gossip section i.e.
+# information about peer nodes. This configuration caps the maximum number of gossip information to carry in each message.
+#
+# Default value is set to 0 which implies the maximum count to be gossiped is set to 10% of all node count.
+# cluster-ping-message-gossip-max-count 0
+
 ########################## CLUSTER DOCKER/NAT support  ########################
 
 # In certain deployments, cluster node's address discovery fails, because