Add SAFE option to SHUTDOWN to reject shutdown in unsafe situations

src/cluster.h

@@ -92,6 +92,7 @@ int getClusterSize(void);
 int getMyShardSlotCount(void);
 int clusterNodePending(clusterNode *node);
 int clusterNodeIsPrimary(clusterNode *n);
+int clusterNodeIsVotingPrimary(clusterNode *n);
 char **getClusterNodesList(size_t *numnodes);
 char *clusterNodeIp(clusterNode *node, client *c);
 int clusterNodeIsReplica(clusterNode *node);

src/cluster_legacy.c

@@ -130,7 +130,7 @@ void clusterCommandFlushslot(client *c);
 
 /* Only primaries that own slots have voting rights.
  * Returns 1 if the node has voting rights, otherwise returns 0. */
-static inline int clusterNodeIsVotingPrimary(clusterNode *n) {
+int clusterNodeIsVotingPrimary(clusterNode *n) {
     return (n->flags & CLUSTER_NODE_PRIMARY) && n->numslots;
 }
 

src/commands.def

@@ -7917,11 +7917,17 @@ struct COMMAND_ARG SHUTDOWN_abort_selector_save_selector_block_save_selector_Sub
 {MAKE_ARG("save",ARG_TYPE_PURE_TOKEN,-1,"SAVE",NULL,NULL,CMD_ARG_NONE,0,NULL)},
 };
 
+/* SHUTDOWN abort_selector save_selector_block force_safe_selector argument table */
+struct COMMAND_ARG SHUTDOWN_abort_selector_save_selector_block_force_safe_selector_Subargs[] = {
+{MAKE_ARG("force",ARG_TYPE_PURE_TOKEN,-1,"FORCE",NULL,"7.0.0",CMD_ARG_NONE,0,NULL)},
+{MAKE_ARG("safe",ARG_TYPE_PURE_TOKEN,-1,"SAVE",NULL,"9.0.0",CMD_ARG_NONE,0,NULL)},
+};
+
 /* SHUTDOWN abort_selector save_selector_block argument table */
 struct COMMAND_ARG SHUTDOWN_abort_selector_save_selector_block_Subargs[] = {
 {MAKE_ARG("save-selector",ARG_TYPE_ONEOF,-1,NULL,NULL,NULL,CMD_ARG_OPTIONAL,2,NULL),.subargs=SHUTDOWN_abort_selector_save_selector_block_save_selector_Subargs},
 {MAKE_ARG("now",ARG_TYPE_PURE_TOKEN,-1,"NOW",NULL,"7.0.0",CMD_ARG_OPTIONAL,0,NULL)},
-{MAKE_ARG("force",ARG_TYPE_PURE_TOKEN,-1,"FORCE",NULL,"7.0.0",CMD_ARG_OPTIONAL,0,NULL)},
+{MAKE_ARG("force-safe-selector",ARG_TYPE_ONEOF,-1,NULL,NULL,NULL,CMD_ARG_OPTIONAL,2,NULL),.subargs=SHUTDOWN_abort_selector_save_selector_block_force_safe_selector_Subargs},
 };
 
 /* SHUTDOWN abort_selector argument table */

src/commands/shutdown.json

@@ -61,6 +61,13 @@
                                 "token": "FORCE",
                                 "optional": true,
                                 "since": "7.0.0"
+                            },
+                            {
+                                "name": "safe",
+                                "type": "pure-token",
+                                "token": "SAFE",
+                                "optional": true,
+                                "since": "9.0.0"
                             }
                         ]
                     },

src/config.c

@@ -98,11 +98,12 @@ configEnum aof_fsync_enum[] = {
     {NULL, 0}};
 
 configEnum shutdown_on_sig_enum[] = {
-    {"default", 0},
+    {"default", SHUTDOWN_NOFLAGS},
     {"save", SHUTDOWN_SAVE},
     {"nosave", SHUTDOWN_NOSAVE},
     {"now", SHUTDOWN_NOW},
     {"force", SHUTDOWN_FORCE},
+    {"safe", SHUTDOWN_SAFE},
     {NULL, 0}};
 
 configEnum repl_diskless_load_enum[] = {

src/db.c

@@ -1314,7 +1314,7 @@ void typeCommand(client *c) {
     addReplyStatus(c, getObjectTypeName(o));
 }
 
-/* SHUTDOWN [[NOSAVE | SAVE] [NOW] [FORCE] | ABORT] */
+/* SHUTDOWN [[NOSAVE | SAVE] [NOW] [FORCE] [SAFE] | ABORT] */
 void shutdownCommand(client *c) {
     int flags = SHUTDOWN_NOFLAGS;
     int abort = 0;
@@ -1329,6 +1329,8 @@ void shutdownCommand(client *c) {
             flags |= SHUTDOWN_FORCE;
         } else if (!strcasecmp(c->argv[i]->ptr, "abort")) {
             abort = 1;
+        } else if (!strcasecmp(c->argv[i]->ptr, "safe")) {
+            flags |= SHUTDOWN_SAFE;
         } else {
             addReplyErrorObject(c, shared.syntaxerr);
             return;

src/server.c

@@ -4553,6 +4553,7 @@ int finishShutdown(void) {
     int save = server.shutdown_flags & SHUTDOWN_SAVE;
     int nosave = server.shutdown_flags & SHUTDOWN_NOSAVE;
     int force = server.shutdown_flags & SHUTDOWN_FORCE;
+    int safe = server.shutdown_flags & SHUTDOWN_SAFE;
 
     /* Log a warning for each replica that is lagging. */
     listIter replicas_iter;
@@ -4575,6 +4576,17 @@ int finishShutdown(void) {
                   num_replicas);
     }
 
+    if (safe && server.cluster_enabled && clusterNodeIsVotingPrimary(getMyClusterNode())) {
+        if (force) {
+            serverLog(LL_WARNING, "I am a voting primary, shutting down may cause the cluster to down. Exit anyway.");
+        } else {
+            serverLog(LL_WARNING, "I am a voting primary, shutting down may cause the cluster to down, can't exit.");
+            if (server.supervised_mode == SUPERVISED_SYSTEMD)
+                serverCommunicateSystemd("I am a voting primary, shutting down may cause the cluster to down, can't exit.\n");
+            goto error;
+        }
+    }
+
     /* Kill all the Lua debugger forked sessions. */
     ldbKillForkedSessions();
 

src/server.h

@@ -570,12 +570,12 @@ typedef enum {
 #define UNIT_MILLISECONDS 1
 
 /* SHUTDOWN flags */
-#define SHUTDOWN_NOFLAGS 0 /* No flags. */
-#define SHUTDOWN_SAVE 1    /* Force SAVE on SHUTDOWN even if no save \
-                              points are configured. */
-#define SHUTDOWN_NOSAVE 2  /* Don't SAVE on SHUTDOWN. */
-#define SHUTDOWN_NOW 4     /* Don't wait for replicas to catch up. */
-#define SHUTDOWN_FORCE 8   /* Don't let errors prevent shutdown. */
+#define SHUTDOWN_NOFLAGS 0       /* No flags. */
+#define SHUTDOWN_SAVE (1 << 0)   /* Force SAVE on SHUTDOWN even if no save points are configured. */
+#define SHUTDOWN_NOSAVE (1 << 1) /* Don't SAVE on SHUTDOWN. */
+#define SHUTDOWN_NOW (1 << 2)    /* Don't wait for replicas to catch up. */
+#define SHUTDOWN_FORCE (1 << 3)  /* Don't let errors prevent shutdown. */
+#define SHUTDOWN_SAFE (1 << 4)   /* Shutdown only when safe. */
 
 /* Command call flags, see call() function */
 #define CMD_CALL_NONE 0

tests/unit/cluster/shutdown.tcl

@@ -0,0 +1,15 @@
+start_cluster 3 0 {tags {external:skip cluster shutdown}} {
+    test "Test shutdown safe is work" {
+        assert_error {ERR Errors trying to SHUTDOWN*} {R 0 shutdown safe}
+        assert_error {ERR Errors trying to SHUTDOWN*} {R 1 shutdown safe}
+        assert_error {ERR Errors trying to SHUTDOWN*} {R 2 shutdown safe}
+
+        catch {R 0 shutdown safe force}
+        wait_for_condition 1000 50 {
+            [CI 1 cluster_state] eq {fail} &&
+            [CI 2 cluster_state] eq {fail}
+        } else {
+            fail "Cluster doesn't fail"
+        }
+    }
+}

valkey.conf

@@ -1705,6 +1705,11 @@ aof-timestamp-enabled no
 # nosave:   Prevents DB saving operation even if one or more save points are configured.
 # now:      Skips waiting for lagging replicas.
 # force:    Ignores any errors that would normally prevent the server from exiting.
+# safe:     Shutdown only when safe. Note that safe cannot prevent force, in the case of
+#           force, safe will print the relevant logs. The definition of safe may be different
+#           in different modes. Here are the definitions:
+#           1. In cluster mode, it is unsafe to shutdown a primary with slots, and may cause
+#              the cluster to go down.
 #
 # Any combination of values is allowed as long as "save" and "nosave" are not set simultaneously.
 # Example: "nosave force now"