Start failover without waiting for the next cluster cron cycle

src/cluster_legacy.c

@@ -2211,6 +2211,12 @@ void markNodeAsFailingIfNeeded(clusterNode *node) {
      * so here the replica is only helping propagating this status. */
     clusterSendFail(node->name);
     clusterDoBeforeSleep(CLUSTER_TODO_UPDATE_STATE | CLUSTER_TODO_SAVE_CONFIG);
+
+    if (nodeIsReplica(myself) && myself->replicaof == node) {
+        /* Immediately check if the node is our primary node, if so, we can try
+         * to initiate a failover as soon as possible. */
+        clusterDoBeforeSleep(CLUSTER_TODO_HANDLE_FASTER_FAILOVER);
+    }
 }
 
 /* This function is called only if a node is marked as FAIL, but we are able
@@ -3543,6 +3549,12 @@ int clusterProcessPacket(clusterLink *link) {
                     noaddr_node->flags |= CLUSTER_NODE_FAIL;
                     noaddr_node->fail_time = now;
                     clusterSendFail(noaddr_node->name);
+
+                    if (nodeIsReplica(myself) && myself->replicaof == noaddr_node) {
+                        /* Immediately check if the node is our primary node, if so, we can try
+                         * to initiate a failover as soon as possible. */
+                        clusterDoBeforeSleep(CLUSTER_TODO_HANDLE_FASTER_FAILOVER);
+                    }
                 }
                 clusterDoBeforeSleep(CLUSTER_TODO_SAVE_CONFIG | CLUSTER_TODO_UPDATE_STATE);
                 return 0;
@@ -3777,6 +3789,15 @@ int clusterProcessPacket(clusterLink *link) {
                 failing->fail_time = now;
                 failing->flags &= ~CLUSTER_NODE_PFAIL;
                 clusterDoBeforeSleep(CLUSTER_TODO_SAVE_CONFIG | CLUSTER_TODO_UPDATE_STATE);
+
+                if (nodeIsReplica(myself) && myself->replicaof == failing) {
+                    /* Immediately check if the node is our primary node, if so, we can try
+                     * to initiate a failover as soon as possible. We will also try to send
+                     * the FAIL packet in case we trigger the failover immediately but unable
+                     * to get the votes. */
+                    clusterSendFail(failing->name);
+                    clusterDoBeforeSleep(CLUSTER_TODO_HANDLE_FASTER_FAILOVER);
+                }
             }
         } else {
             serverLog(LL_NOTICE, "Ignoring FAIL message from unknown node %.40s about %.40s", hdr->sender,
@@ -4963,7 +4984,6 @@ void clusterHandleReplicaFailover(void) {
      * elapsed, we can setup a new one. */
     if (auth_age > auth_retry_time) {
         server.cluster->failover_auth_time = now +
-                                             500 +           /* Fixed delay of 500 milliseconds, let FAIL msg propagate. */
                                              random() % 500; /* Random delay between 0 and 500 milliseconds. */
         server.cluster->failover_auth_count = 0;
         server.cluster->failover_auth_sent = 0;
@@ -4984,6 +5004,11 @@ void clusterHandleReplicaFailover(void) {
             /* Reset auth_age since it is outdated now and we can bypass the auth_timeout
              * check in the next state and start the election ASAP. */
             auth_age = 0;
+        } else if (server.cluster->failover_auth_time == now) {
+            /* If we happen to initiate a failover immediately, reset auth_age since it is
+             * outdated now and we can bypass the auth_timeout check in the next state and
+             * start the election ASAP. */
+            auth_age = 0;
         }
         serverLog(LL_NOTICE,
                   "Start of election delayed for %lld milliseconds "
@@ -5573,7 +5598,8 @@ void clusterBeforeSleep(void) {
         }
     } else if (flags & CLUSTER_TODO_HANDLE_FAILOVER) {
         /* Handle failover, this is needed when it is likely that there is already
-         * the quorum from primaries in order to react fast. */
+         * the quorum from primaries in order to react fast. Or when we determine
+         * that we can proceed with the failover. */
         clusterHandleReplicaFailover();
     }
 
@@ -5592,6 +5618,12 @@ void clusterBeforeSleep(void) {
          * regular ping. */
         clusterBroadcastPong(CLUSTER_BROADCAST_ALL);
     }
+
+    if (flags & CLUSTER_TODO_HANDLE_FASTER_FAILOVER) {
+        /* Handle faster failover, this goes after all the todos to check if we
+         * can initiate a fast failover. */
+        clusterHandleReplicaFailover();
+    }
 }
 
 void clusterDoBeforeSleep(int flags) {

src/cluster_legacy.h

@@ -25,6 +25,7 @@
 #define CLUSTER_TODO_FSYNC_CONFIG (1 << 3)
 #define CLUSTER_TODO_HANDLE_MANUALFAILOVER (1 << 4)
 #define CLUSTER_TODO_BROADCAST_ALL (1 << 5)
+#define CLUSTER_TODO_HANDLE_FASTER_FAILOVER (1 << 6)
 
 /* clusterLink encapsulates everything needed to talk with a remote node. */
 typedef struct clusterLink {