fix: preserve HMS table properties during commits

This change fixes HMS-specific table properties being lost during
commits by merging parameters instead of replacing them.

Fixes: #2926

Author: bharos

pyiceberg/catalog/hive.py

@@ -551,11 +551,22 @@ def commit_table(
 
                 if hive_table and current_table:
                     # Table exists, update it.
-                    hive_table.parameters = _construct_parameters(
+                    new_parameters = _construct_parameters(
                         metadata_location=updated_staged_table.metadata_location,
                         previous_metadata_location=current_table.metadata_location,
                         metadata_properties=updated_staged_table.properties,
                     )
+
+                    # Detect properties that were removed from Iceberg metadata
+                    old_iceberg_keys = set(current_table.properties.keys())
+                    new_iceberg_keys = set(updated_staged_table.properties.keys())
+                    removed_keys = old_iceberg_keys - new_iceberg_keys
+
+                    # Start with current HMS parameters, remove deleted Iceberg properties, then update with new ones
+                    updated_parameters = {k: v for k, v in hive_table.parameters.items() if k not in removed_keys}
+                    updated_parameters.update(new_parameters)
+                    hive_table.parameters = updated_parameters
+
                     # Update hive's schema and properties
                     hive_table.sd = _construct_hive_storage_descriptor(
                         updated_staged_table.schema(),

tests/integration/test_reads.py

@@ -135,6 +135,44 @@ def test_hive_properties(catalog: Catalog) -> None:
         assert hive_table.parameters.get("abc") is None
 
 
+@pytest.mark.integration
+@pytest.mark.parametrize("catalog", [pytest.lazy_fixture("session_catalog_hive")])
+def test_hive_preserves_hms_specific_properties(catalog: Catalog) -> None:
+    """Test that HMS-specific table properties are preserved during table commits.
+
+    This verifies that HMS-specific properties that are not managed by Iceberg
+    are preserved during commits, rather than being lost.
+
+    Regression test for: https://github.com/apache/iceberg-python/issues/2926
+    """
+    table = create_table(catalog)
+    hive_client: _HiveClient = _HiveClient(catalog.properties["uri"])
+    with hive_client as open_client:
+        hive_table = open_client.get_table(*TABLE_NAME)
+        # Add HMS-specific properties that aren't managed by Iceberg
+        hive_table.parameters["table_category"] = "production"
+        hive_table.parameters["data_owner"] = "data_team"
+        open_client.alter_table(TABLE_NAME[0], TABLE_NAME[1], hive_table)
+
+    with hive_client as open_client:
+        hive_table = open_client.get_table(*TABLE_NAME)
+        assert hive_table.parameters.get("table_category") == "production"
+        assert hive_table.parameters.get("data_owner") == "data_team"
+
+    table.transaction().set_properties({"iceberg_property": "new_value"}).commit_transaction()
+
+    # Verify that HMS-specific properties are STILL present after commit
+    with hive_client as open_client:
+        hive_table = open_client.get_table(*TABLE_NAME)
+        # HMS-specific properties should be preserved
+        assert hive_table.parameters.get("table_category") == "production", (
+            "HMS property 'table_category' was lost during commit!"
+        )
+        assert hive_table.parameters.get("data_owner") == "data_team", "HMS property 'data_owner' was lost during commit!"
+        # Iceberg properties should also be present
+        assert hive_table.parameters.get("iceberg_property") == "new_value"
+
+
 @pytest.mark.integration
 @pytest.mark.parametrize("catalog", [pytest.lazy_fixture("session_catalog_hive"), pytest.lazy_fixture("session_catalog")])
 def test_table_properties_dict(catalog: Catalog) -> None: