diff --git a/src/graph/executor/query/TraverseExecutor.h b/src/graph/executor/query/TraverseExecutor.h
index 7fe4e0a3b61..424945eec5c 100644
--- a/src/graph/executor/query/TraverseExecutor.h
+++ b/src/graph/executor/query/TraverseExecutor.h
@@ -14,7 +14,11 @@
 // invoke the getNeighbors interface, according to the number of times specified by the user,
 // and assemble the result into paths
 //
-//  The definition of path is : array of vertex and edges
+//  Path is an array of vertex and edges physically.
+//  Its definition is a trail, in which all edges are distinct. It's different to a walk
+//  which allows duplicated vertices and edges, and a path where all vertices and edges
+//  are distinct.
+//
 //  Eg a->b->c. path is [Vertex(a), [Edge(a->b), Vertex(b), Edge(b->c), Vertex(c)]]
 //  the purpose is to extract the path by pathBuildExpression
 // `resDs_` : keep result dataSet
diff --git a/src/storage/exec/GetNeighborsNode.h b/src/storage/exec/GetNeighborsNode.h
index ee9719b7458..6fe2b2ecd07 100644
--- a/src/storage/exec/GetNeighborsNode.h
+++ b/src/storage/exec/GetNeighborsNode.h
@@ -116,6 +116,9 @@ class GetNeighborsNode : public QueryNode<VertexID> {
         return nebula::cpp2::ErrorCode::SUCCEEDED;
       }
       auto key = upstream_->key();
+      if (isDuplicatedSelfReflectiveEdge(key)) {
+        continue;
+      }
       auto reader = upstream_->reader();
       auto props = context_->props_;
       auto columnIdx = context_->columnIdx_;
@@ -138,6 +141,24 @@ class GetNeighborsNode : public QueryNode<VertexID> {
     return nebula::cpp2::ErrorCode::SUCCEEDED;
   }
 
+  bool isDuplicatedSelfReflectiveEdge(const folly::StringPiece& key) {
+    folly::StringPiece srcID = NebulaKeyUtils::getSrcId(context_->vIdLen(), key);
+    folly::StringPiece dstID = NebulaKeyUtils::getDstId(context_->vIdLen(), key);
+    if (srcID == dstID) {
+      // self-reflective edge
+      std::string rank = std::to_string(NebulaKeyUtils::getRank(context_->vIdLen(), key));
+      auto edgeType = NebulaKeyUtils::getEdgeType(context_->vIdLen(), key);
+      edgeType = edgeType > 0 ? edgeType : -edgeType;
+      std::string type = std::to_string(edgeType);
+      std::string localKey = type + rank + srcID.str();
+      if (!visitedSelfReflectiveEdges_.insert(localKey).second) {
+        return true;
+      }
+    }
+    return false;
+  }
+
+  std::unordered_set<std::string> visitedSelfReflectiveEdges_;
   RuntimeContext* context_;
   IterateNode<VertexID>* hashJoinNode_;
   IterateNode<VertexID>* upstream_;
diff --git a/tests/tck/features/match/SelfReflectiveEdges.feature b/tests/tck/features/match/SelfReflectiveEdges.feature
new file mode 100644
index 00000000000..191ef8238e6
--- /dev/null
+++ b/tests/tck/features/match/SelfReflectiveEdges.feature
@@ -0,0 +1,40 @@
+# Copyright (c) 2022 vesoft inc. All rights reserved.
+#
+# This source code is licensed under Apache 2.0 License.
+Feature: Matches on self-reflective edges
+
+  Scenario: no duplicated self reflective edges
+    Given a graph with space named "nba"
+    And having executed:
+      """
+      insert vertex player (name, age) values "Hades":("Hades", 99999);
+      insert vertex team (name) values "Underworld":("Underworld");
+      insert edge like (likeness) values "Hades"->"Hades":(3000);
+      insert edge teammate (start_year, end_year) values "Hades"->"Hades":(3000, 3000);
+      insert edge serve (start_year, end_year) values "Hades"->"Underworld":(0, 99999);
+      """
+    When executing query:
+      """
+      MATCH x = (n0)-[e1]->(n1)-[e2]-(n0) WHERE id(n0) == "Hades" and id(n1) == "Hades" return e1, e2
+      """
+    Then the result should be, in any order:
+      | e1                                                                 | e2                                                                 |
+      | [:teammate "Hades"->"Hades" @0 {end_year: 3000, start_year: 3000}] | [:like "Hades"->"Hades" @0 {likeness: 3000}]                       |
+      | [:like "Hades"->"Hades" @0 {likeness: 3000}]                       | [:teammate "Hades"->"Hades" @0 {end_year: 3000, start_year: 3000}] |
+    When executing query:
+      """
+      MATCH x = (n0)-[e1]->(n1)<-[e2]-(n0) WHERE id(n0) == "Hades" and id(n1) == "Hades" return e1, e2
+      """
+    Then the result should be, in any order:
+      | e1                                                                 | e2                                                                 |
+      | [:teammate "Hades"->"Hades" @0 {end_year: 3000, start_year: 3000}] | [:like "Hades"->"Hades" @0 {likeness: 3000}]                       |
+      | [:like "Hades"->"Hades" @0 {likeness: 3000}]                       | [:teammate "Hades"->"Hades" @0 {end_year: 3000, start_year: 3000}] |
+    When executing query:
+      """
+      DELETE EDGE serve "Hades"->"Underworld";
+      DELETE EDGE teammate "Hades"->"Hades";
+      DELETE EDGE like "Hades"->"Hades";
+      DELETE VERTEX "Underworld";
+      DELETE VERTEX "Hades";
+      """
+    Then the execution should be successful