diff --git a/src/query/management/src/warehouse/warehouse_mgr.rs b/src/query/management/src/warehouse/warehouse_mgr.rs
index b02bcbcb376ac..13a2ea51065d8 100644
--- a/src/query/management/src/warehouse/warehouse_mgr.rs
+++ b/src/query/management/src/warehouse/warehouse_mgr.rs
@@ -1015,6 +1015,11 @@ impl WarehouseApi for WarehouseMgr {
                         value: Some(seq_v), ..
                     })),
             }) => Ok(seq_v.seq),
+            // compatibility
+            // After network fail, nodes may become expired due to failed heartbeats.
+            // For system-managed nodes, this situation has already been handled in resolve_conflicts.
+            // For self-managed nodes, we need to return seq = 0 so that the next heartbeat can proceed normally.
+            _ if matches!(node.node_type, NodeType::SelfManaged) => Ok(0),
             _ => Err(ErrorCode::MetaServiceError("Heartbeat node info failure.")),
         }
     }
diff --git a/src/query/management/tests/it/warehouse.rs b/src/query/management/tests/it/warehouse.rs
index 17632921490e2..aabbe97e920d3 100644
--- a/src/query/management/tests/it/warehouse.rs
+++ b/src/query/management/tests/it/warehouse.rs
@@ -210,6 +210,32 @@ async fn test_drop_self_managed_warehouse() -> Result<()> {
     Ok(())
 }
 
+#[tokio::test(flavor = "multi_thread", worker_threads = 1)]
+async fn test_heartbeat_not_exists_self_managed_node() -> Result<()> {
+    let (_kv, warehouse_manager, _nodes) = nodes(Duration::from_mins(60), 0).await?;
+
+    let node_info = self_managed_node("test_node");
+
+    // heartbeat not exists node info
+    let mut heartbeat_node = node_info.clone();
+    let seq = warehouse_manager
+        .heartbeat_node(&mut heartbeat_node, 34234)
+        .await?;
+
+    assert_eq!(seq, 0);
+    assert_eq!(heartbeat_node, node_info);
+
+    let mut heartbeat_node = node_info.clone();
+    let seq = warehouse_manager
+        .heartbeat_node(&mut heartbeat_node, seq)
+        .await?;
+
+    assert_ne!(seq, 0);
+    assert_eq!(heartbeat_node, node_info);
+
+    Ok(())
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 1)]
 async fn test_successfully_heartbeat_self_managed_node() -> Result<()> {
     let (kv, warehouse_manager, _nodes) = nodes(Duration::from_mins(60), 0).await?;
diff --git a/src/query/service/src/clusters/cluster.rs b/src/query/service/src/clusters/cluster.rs
index acfbe065fec72..5ba2fa91125c6 100644
--- a/src/query/service/src/clusters/cluster.rs
+++ b/src/query/service/src/clusters/cluster.rs
@@ -323,8 +323,15 @@ impl ClusterDiscovery {
                     cluster_nodes.len() as f64,
                 );
 
-                let res = Cluster::create(res, self.local_id.clone());
-                Ok(res)
+                // compatibility, for self-managed nodes, we allow queries to continue executing even when the heartbeat fails.
+                if cluster_nodes.is_empty() && !config.query.cluster_id.is_empty() {
+                    let mut cluster = Cluster::empty();
+                    let mut_cluster = Arc::get_mut(&mut cluster).unwrap();
+                    mut_cluster.local_id = self.local_id.clone();
+                    return Ok(cluster);
+                }
+
+                Ok(Cluster::create(res, self.local_id.clone()))
             }
         }
     }
diff --git a/src/query/service/tests/it/clusters.rs b/src/query/service/tests/it/clusters.rs
index 96b31b3e5899c..e7c19d4d893c1 100644
--- a/src/query/service/tests/it/clusters.rs
+++ b/src/query/service/tests/it/clusters.rs
@@ -19,6 +19,25 @@ use databend_query::clusters::ClusterHelper;
 use databend_query::test_kits::*;
 use pretty_assertions::assert_eq;
 
+#[tokio::test(flavor = "multi_thread", worker_threads = 1)]
+async fn test_empty_cluster_discovery() -> Result<()> {
+    let _guard = TestFixture::setup().await?;
+
+    let config = ConfigBuilder::create().build();
+
+    let metastore = ClusterDiscovery::create_meta_client(&config).await?;
+    let cluster_discovery = ClusterDiscovery::try_create(&config, metastore.clone()).await?;
+
+    let discover_cluster = cluster_discovery.discover(&config).await?;
+
+    let discover_cluster_nodes = discover_cluster.get_nodes();
+    assert_eq!(discover_cluster_nodes.len(), 0);
+    assert!(discover_cluster.is_empty());
+    assert!(!discover_cluster.unassign);
+
+    Ok(())
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 1)]
 async fn test_single_cluster_discovery() -> Result<()> {
     let config = ConfigBuilder::create().build();