diff --git a/src/sql/engine/join/ob_merge_join_op.cpp b/src/sql/engine/join/ob_merge_join_op.cpp
index 2a7351fd92..d8e5a731b8 100644
--- a/src/sql/engine/join/ob_merge_join_op.cpp
+++ b/src/sql/engine/join/ob_merge_join_op.cpp
@@ -875,6 +875,36 @@ int ObMergeJoinOp::set_is_match(const int64_t idx, const bool is_match)
   return ret;
 }
 
+int ObMergeJoinOp::update_store_mem_bound(ObRADatumStore *left, ObRADatumStore *right)
+{
+  int ret = OB_SUCCESS;
+  bool updated = false;
+  ObMergeJoinMemChecker checker(get_total_rows_in_datum_store());
+  if (OB_FAIL(sql_mem_processor_.update_max_available_mem_size_periodically(
+              &mem_context_->get_malloc_allocator(),
+              checker,
+              updated))) {
+    LOG_WARN("failed to update max available memory size periodically", K(ret));
+  } else {
+    int64_t t_mem_bound = sql_mem_processor_.get_mem_bound();
+    int64_t l_mem_bound = static_cast<int64_t>(left_mem_bound_ratio_ * t_mem_bound);
+    int64_t r_mem_bound = t_mem_bound - l_mem_bound;
+    if (OB_UNLIKELY(t_mem_bound < 0) || OB_UNLIKELY(l_mem_bound < 0) || OB_UNLIKELY(r_mem_bound < 0)) {
+      ret = OB_ERR_UNEXPECTED;
+      LOG_WARN("unexpected negative mem bound", K(ret), K(t_mem_bound), K(l_mem_bound),
+        K(r_mem_bound), K(left_mem_bound_ratio_));
+    } else {
+      if (OB_NOT_NULL(left)) {
+        left->set_mem_limit(l_mem_bound);
+      }
+      if (OB_NOT_NULL(right)) {
+        right->set_mem_limit(r_mem_bound);
+      }
+    }
+  }
+  return ret;
+}
+
 int ObMergeJoinOp::process_dump()
 {
   int ret = OB_SUCCESS;
@@ -1107,6 +1137,11 @@ int ObMergeJoinOp::ChildBatchFetcher::get_next_small_group(int64_t &cmp_res)
   bool enough_datums = merge_join_op_.has_enough_datums();
   JoinRowList row_list(datum_store_.get_row_cnt());
   ObEvalCtx::BatchInfoScopeGuard guard(merge_join_op_.eval_ctx_);
+  if (need_store_unmatch && OB_FAIL(merge_join_op_.update_store_mem_bound(
+                                is_left ? &datum_store_ : nullptr,
+                                is_left ? nullptr : &datum_store_))) {
+    LOG_WARN("failed to update max available memory size periodically", K(ret));
+  }
   while (OB_SUCC(ret) && !all_batch_finished && !greater_found && !enough_datums) {
     if (need_store_unmatch && OB_LIKELY(cur_idx_ < brs_.size_)) {
       ObRADatumStore::StoredRow *stored_row = NULL;
@@ -1366,18 +1401,8 @@ int ObMergeJoinOp::iterate_both_chidren(ObEvalCtx::BatchInfoScopeGuard &guard)
   JoinRowList r_row_list(r_store->get_row_cnt());
   ObRADatumStore::StoredRow *l_row = NULL;
   ObRADatumStore::StoredRow *r_row = NULL;
-  bool updated = false;
-  if (OB_FAIL(sql_mem_processor_.update_max_available_mem_size_periodically(
-              &mem_context_->get_malloc_allocator(),
-              [&](int64_t cur_cnt) { return get_total_rows_in_datum_store() > cur_cnt; },
-              updated))) {
+  if (OB_FAIL(update_store_mem_bound(l_store, r_store))) {
     LOG_WARN("failed to update max available memory size periodically", K(ret));
-  }
-  int64_t t_mem_bound = sql_mem_processor_.get_mem_bound();
-  int64_t l_mem_bound = static_cast<int64_t>(left_mem_bound_ratio_ * t_mem_bound);
-  int64_t r_mem_bound = 0; // will be set to the remaining memory size later.
-  l_store->set_mem_limit(l_mem_bound);
-  if (OB_FAIL(ret)) {
   } else if (OB_FAIL(store_group_first_row(left_brs_fetcher_, l_row_list, l_row, guard))) {
     LOG_WARN("store left group first row failed", K(ret));
   } else if (OB_FAIL(store_group_first_row(right_brs_fetcher_, r_row_list, r_row, guard))) {
@@ -1386,11 +1411,7 @@ int ObMergeJoinOp::iterate_both_chidren(ObEvalCtx::BatchInfoScopeGuard &guard)
                                                                    MY_SPEC.is_left_unique_,
                                                                    l_row)))) {
     LOG_WARN("get left next group failed", K(ret));
-  } else if (OB_UNLIKELY((r_mem_bound = t_mem_bound - l_store->get_mem_hold()) < 0)) {
-    ret = OB_ERR_UNEXPECTED;
-    LOG_WARN("unexpected negative mem bound", K(ret), K(t_mem_bound), K(l_mem_bound),
-                                              K(r_mem_bound), K(l_store->get_mem_hold()));
-  } else if (FALSE_IT(r_store->set_mem_limit(r_mem_bound))) {
+
   } else if (OB_FAIL((right_brs_fetcher_.get_next_equal_group<false>(r_row_list, l_row, false,
                                                                      r_row)))) {
     LOG_WARN("get right next group failed", K(ret));
diff --git a/src/sql/engine/join/ob_merge_join_op.h b/src/sql/engine/join/ob_merge_join_op.h
index 6fa8dac98e..087dec6ab4 100644
--- a/src/sql/engine/join/ob_merge_join_op.h
+++ b/src/sql/engine/join/ob_merge_join_op.h
@@ -23,6 +23,18 @@ namespace oceanbase
 {
 namespace sql
 {
+class ObMergeJoinMemChecker
+{
+public:
+  ObMergeJoinMemChecker(int64_t row_cnt): cur_row_cnt_(row_cnt)
+  {}
+  bool operator()(int64_t max_row_count)
+  {
+    return cur_row_cnt_ > max_row_count;
+  }
+  int64_t cur_row_cnt_;
+};
+
 class ObMergeJoinSpec: public ObJoinSpec
 {
   OB_UNIS_VERSION_V(1);
@@ -490,6 +502,7 @@ private:
   inline bool is_match(const int64_t idx) { return rj_match_vec_->at(idx); }
   // this function only used for non-vectorized code patch
   int process_dump();
+  int update_store_mem_bound(ObRADatumStore *left, ObRADatumStore *right);
   typedef int (ObMergeJoinOp::*state_operation_func_type)();
   typedef int (ObMergeJoinOp::*state_function_func_type)();
 private: