diff --git a/src/Interpreters/ActionsDAG.cpp b/src/Interpreters/ActionsDAG.cpp
index 04dee2ed6e6..af1f2db540b 100644
--- a/src/Interpreters/ActionsDAG.cpp
+++ b/src/Interpreters/ActionsDAG.cpp
@@ -464,12 +464,19 @@ void ActionsDAG::removeUnusedActions(const Names & required_names, bool allow_re
 
 void ActionsDAG::removeUnusedActions(bool allow_remove_inputs, bool allow_constant_folding)
 {
-    std::unordered_set<const Node *> visited_nodes;
     std::unordered_set<const Node *> used_inputs;
-    std::stack<Node *> stack;
+    if (!allow_remove_inputs)
+    {
+        for (const auto * input : inputs)
+            used_inputs.insert(input);
+    }
+    removeUnusedActions(used_inputs, allow_constant_folding);
+}
 
-    for (const auto * input : inputs)
-        used_inputs.insert(input);
+void ActionsDAG::removeUnusedActions(const std::unordered_set<const Node *> & used_inputs, bool allow_constant_folding)
+{
+    std::unordered_set<const Node *> visited_nodes;
+    std::stack<Node *> stack;
 
     for (const auto * node : outputs)
     {
@@ -488,7 +495,7 @@ void ActionsDAG::removeUnusedActions(bool allow_remove_inputs, bool allow_consta
             stack.push(&node);
         }
 
-        if (node.type == ActionType::INPUT && !allow_remove_inputs && used_inputs.contains(&node))
+        if (node.type == ActionType::INPUT && used_inputs.contains(&node))
             visited_nodes.insert(&node);
     }
 
@@ -2187,14 +2194,6 @@ ActionsDAGPtr ActionsDAG::cloneActionsForFilterPushDown(
         {
             /// If filter column is not needed, remove it from output nodes.
             std::erase_if(outputs, [&](const Node * node) { return node == predicate; });
-
-            /// At the very end of this method we'll call removeUnusedActions() with allow_remove_inputs=false,
-            /// so we need to manually remove predicate if it is an input node.
-            if (predicate->type == ActionType::INPUT)
-            {
-                std::erase_if(inputs, [&](const Node * node) { return node == predicate; });
-                nodes.remove_if([&](const Node & node) { return &node == predicate; });
-            }
         }
         else
         {
@@ -2261,7 +2260,15 @@ ActionsDAGPtr ActionsDAG::cloneActionsForFilterPushDown(
         }
     }
 
-    removeUnusedActions(false);
+    std::unordered_set<const Node *> used_inputs;
+    for (const auto * input : inputs)
+    {
+        if (can_remove_filter && input == predicate)
+            continue;
+        used_inputs.insert(input);
+    }
+
+    removeUnusedActions(used_inputs);
     return actions;
 }
 
diff --git a/src/Interpreters/ActionsDAG.h b/src/Interpreters/ActionsDAG.h
index 48ed03d7347..e6dff1ffb0b 100644
--- a/src/Interpreters/ActionsDAG.h
+++ b/src/Interpreters/ActionsDAG.h
@@ -182,6 +182,9 @@ public:
     /// Remove actions that are not needed to compute output nodes
     void removeUnusedActions(bool allow_remove_inputs = true, bool allow_constant_folding = true);
 
+    /// Remove actions that are not needed to compute output nodes. Keep inputs from used_inputs.
+    void removeUnusedActions(const std::unordered_set<const Node *> & used_inputs, bool allow_constant_folding = true);
+
     /// Remove actions that are not needed to compute output nodes with required names
     void removeUnusedActions(const Names & required_names, bool allow_remove_inputs = true, bool allow_constant_folding = true);
 
diff --git a/tests/queries/0_stateless/02907_filter_pushdown_crash.reference b/tests/queries/0_stateless/02907_filter_pushdown_crash.reference
new file mode 100644
index 00000000000..e69de29bb2d
diff --git a/tests/queries/0_stateless/02907_filter_pushdown_crash.sql b/tests/queries/0_stateless/02907_filter_pushdown_crash.sql
new file mode 100644
index 00000000000..eb881823f27
--- /dev/null
+++ b/tests/queries/0_stateless/02907_filter_pushdown_crash.sql
@@ -0,0 +1,18 @@
+DROP TABLE IF EXISTS t1;
+DROP TABLE IF EXISTS t2;
+
+CREATE TABLE t1 (key UInt8) ENGINE = MergeTree ORDER BY key;
+INSERT INTO t1 VALUES (1),(2);
+
+CREATE TABLE t2 (key UInt32) ENGINE = MergeTree ORDER BY key;
+INSERT INTO t2 VALUES (1),(2);
+
+SELECT a FROM ( SELECT key + 1 as a, key FROM t1 GROUP BY key ) WHERE key FORMAT Null;
+
+SET join_algorithm = 'full_sorting_merge';
+SET max_rows_in_set_to_optimize_join = 0;
+
+SELECT key FROM ( SELECT key FROM t1 GROUP BY key ) t1 JOIN (SELECT key FROM t2) t2 ON t1.key = t2.key WHERE key FORMAT Null;
+
+DROP TABLE IF EXISTS t1;
+DROP TABLE IF EXISTS t2;