Fix: Add method to merge conversation stats (#10667)

Co-authored-by: openhands <openhands@all-hands.dev>

tests/unit/test_conversation_stats.py

@@ -490,6 +490,117 @@ def test_save_and_restore_workflow(mock_file_store):
         assert llm.metrics.accumulated_token_usage.completion_tokens == 50
 
 
+def test_merge_conversation_stats_success_non_overlapping(mock_file_store):
+    """Merging two ConversationStats combines only restored metrics. Active metrics
+    (service_to_metrics) are not merged; if present, an error is logged but
+    execution continues. Incoming restored metrics overwrite duplicates.
+    """
+    stats_a = ConversationStats(
+        file_store=mock_file_store, conversation_id='conv-merge-a', user_id='user-x'
+    )
+    stats_b = ConversationStats(
+        file_store=mock_file_store, conversation_id='conv-merge-b', user_id='user-x'
+    )
+
+    # Self active + restored
+    m_a_active = Metrics(model_name='model-a')
+    m_a_active.add_cost(0.1)
+    m_a_restored = Metrics(model_name='model-a')
+    m_a_restored.add_cost(0.2)
+    stats_a.service_to_metrics['a-active'] = m_a_active
+    stats_a.restored_metrics = {'a-restored': m_a_restored}
+
+    # Other active + restored
+    m_b_active = Metrics(model_name='model-b')
+    m_b_active.add_cost(0.3)
+    m_b_restored = Metrics(model_name='model-b')
+    m_b_restored.add_cost(0.4)
+    stats_b.service_to_metrics['b-active'] = m_b_active
+    stats_b.restored_metrics = {'b-restored': m_b_restored}
+
+    # Merge B into A
+    stats_a.merge_and_save(stats_b)
+
+    # Active metrics should not be merged; only self's active metrics remain
+    assert set(stats_a.service_to_metrics.keys()) == {
+        'a-active',
+    }
+
+    # Restored metrics from both sides should be in A's restored_metrics
+    assert set(stats_a.restored_metrics.keys()) == {
+        'a-restored',
+        'b-restored',
+    }
+
+    # The exact Metrics objects should be present (no copies)
+    assert stats_a.service_to_metrics['a-active'] is m_a_active
+    assert stats_a.restored_metrics['a-restored'] is m_a_restored
+    assert stats_a.restored_metrics['b-restored'] is m_b_restored
+
+    # Merge triggers a save; confirm the saved blob decodes to expected keys
+    # The save_metrics method combines both service_to_metrics and restored_metrics
+    encoded = mock_file_store.read(stats_a.metrics_path)
+    pickled = base64.b64decode(encoded)
+    restored_dict = pickle.loads(pickled)
+    assert set(restored_dict.keys()) == {
+        'a-active',
+        'a-restored',
+        'b-restored',
+    }
+
+
+@pytest.mark.parametrize(
+    'self_side,other_side',
+    [
+        ('active', 'active'),
+        ('restored', 'active'),
+        ('active', 'restored'),
+        ('restored', 'restored'),
+    ],
+)
+def test_merge_conversation_stats_duplicates_overwrite_and_log_errors(
+    mock_file_store, self_side, other_side
+):
+    stats_a = ConversationStats(
+        file_store=mock_file_store, conversation_id='conv-merge-a', user_id='user-x'
+    )
+    stats_b = ConversationStats(
+        file_store=mock_file_store, conversation_id='conv-merge-b', user_id='user-x'
+    )
+
+    # Place the same service id on the specified sides
+    dupe_id = 'dupe-service'
+    m1 = Metrics(model_name='m')
+    m1.add_cost(0.1)  # ensure not dropped
+    m2 = Metrics(model_name='m')
+    m2.add_cost(0.2)  # ensure not dropped
+
+    if self_side == 'active':
+        stats_a.service_to_metrics[dupe_id] = m1
+    else:
+        stats_a.restored_metrics[dupe_id] = m1
+
+    if other_side == 'active':
+        stats_b.service_to_metrics[dupe_id] = m2
+    else:
+        stats_b.restored_metrics[dupe_id] = m2
+
+    # Perform merge; should not raise and should log error internally if active metrics present
+    stats_a.merge_and_save(stats_b)
+
+    # Only restored metrics are merged; duplicates are allowed with incoming overwriting
+    if other_side == 'restored':
+        assert dupe_id in stats_a.restored_metrics
+        assert stats_a.restored_metrics[dupe_id] is m2  # incoming overwrites existing
+    else:
+        # No restored metric came from incoming; existing restored stays as-is
+        if self_side == 'restored':
+            assert dupe_id in stats_a.restored_metrics
+            assert stats_a.restored_metrics[dupe_id] is m1
+        else:
+            assert dupe_id not in stats_a.restored_metrics
+
+
 def test_save_metrics_preserves_restored_metrics_fix(mock_file_store):
     """Test that save_metrics correctly preserves restored metrics for unregistered services."""
     conversation_id = 'test-conversation-id'