import numbers import numpy as np import pytest from base.client_v2_base import TestMilvusClientV2Base from common import common_func as cf from common import common_type as ct from common.common_type import CaseLabel, CheckTasks from utils.util_log import test_log as log from utils.util_pymilvus import DataType prefix = "alter" default_vector_field_name = "vector" default_primary_key_field_name = "id" default_string_field_name = "varchar" default_float_field_name = "float" default_new_field_name = "field_new" default_dynamic_field_name = "dynamic_field" exp_res = "exp_res" default_nb = 20 default_dim = 128 default_limit = 10 default_warmup_dim = 128 default_warmup_nb = ct.default_nb class TestMilvusClientAlterIndex(TestMilvusClientV2Base): @pytest.mark.tags(CaseLabel.L0) def test_milvus_client_alter_index_default(self): """ target: test alter index method: 1. alter index after load verify alter fail 2. alter index after release verify alter successfully 3. drop index properties after load verify drop fail 4. drop index properties after release verify drop successfully expected: alter successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() self.create_collection(client, collection_name, ct.default_dim, consistency_level="Strong") idx_names, _ = self.list_indexes(client, collection_name, field_name=default_vector_field_name) self.load_collection(client, collection_name) res1 = self.describe_index(client, collection_name, index_name=idx_names[0])[0] assert res1.get("mmap.enabled", None) is None error = { ct.err_code: 104, ct.err_msg: f"can't alter index on loaded collection, " f"please release the collection first: collection already loaded[collection={collection_name}]", } # 1. alter index after load self.alter_index_properties( client, collection_name, idx_names[0], properties={"mmap.enabled": True}, check_task=CheckTasks.err_res, check_items=error, ) self.drop_index_properties( client, collection_name, idx_names[0], property_keys=["mmap.enabled"], check_task=CheckTasks.err_res, check_items=error, ) self.release_collection(client, collection_name) # 2. alter index after release self.alter_index_properties(client, collection_name, idx_names[0], properties={"mmap.enabled": True}) res2 = self.describe_index(client, collection_name, index_name=idx_names[0])[0] assert res2.get("mmap.enabled", None) == "True" self.drop_index_properties(client, collection_name, idx_names[0], property_keys=["mmap.enabled"]) res3 = self.describe_index(client, collection_name, index_name=idx_names[0])[0] assert res3.get("mmap.enabled", None) is None @pytest.mark.tags(CaseLabel.L1) def test_milvus_client_alter_index_unsupported_properties(self): """ target: test alter index with unsupported properties method: 1. alter index with unsupported properties expected: raise exception """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. create collection schema = self.create_schema(client, enable_dynamic_field=False)[0] dim = 32 pk_field_name = "id_string" vector_field_name = "embeddings" str_field_name = "title" max_length = 16 schema.add_field(pk_field_name, DataType.VARCHAR, max_length=max_length, is_primary=True, auto_id=False) schema.add_field(vector_field_name, DataType.FLOAT_VECTOR, dim=dim, mmap_enabled=True) schema.add_field(str_field_name, DataType.VARCHAR, max_length=max_length, mmap_enabled=True) index_params = self.prepare_index_params(client)[0] index_params.add_index( field_name=vector_field_name, metric_type="COSINE", index_type="HNSW", params={"M": 16, "efConstruction": 100, "mmap.enabled": True}, ) index_params.add_index(field_name=str_field_name) self.create_collection( client, collection_name, schema=schema, index_params=index_params, properties={"mmap.enabled": True} ) self.describe_collection( client, collection_name, check_task=CheckTasks.check_collection_fields_properties, check_items={ str_field_name: {"max_length": max_length, "mmap_enabled": True}, vector_field_name: {"mmap_enabled": True}, "properties": {"mmap.enabled": "False"}, }, ) res = self.describe_index(client, collection_name, index_name=vector_field_name)[0] assert res.get("mmap.enabled", None) == "True" self.release_collection(client, collection_name) properties = self.describe_index(client, collection_name, index_name=vector_field_name)[0] for p in properties.items(): if p[0] not in ["mmap.enabled"]: log.debug(f"try to alter index property: {p[0]}") error = {ct.err_code: 1, ct.err_msg: f"{p[0]} is not a configable index property"} new_value = p[1] + 1 if isinstance(p[1], numbers.Number) else "new_value" self.alter_index_properties( client, collection_name, vector_field_name, properties={p[0]: new_value}, check_task=CheckTasks.err_res, check_items=error, ) @pytest.mark.tags(CaseLabel.L1) def test_milvus_client_alter_index_unsupported_value(self): """ target: test alter index with unsupported properties method: 1. alter index with unsupported properties expected: raise exception """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() self.create_collection(client, collection_name, ct.default_dim, consistency_level="Strong") idx_names, _ = self.list_indexes(client, collection_name, field_name=default_vector_field_name) self.release_collection(client, collection_name) res1 = self.describe_index(client, collection_name, index_name=idx_names[0])[0] assert res1.get("mmap.enabled", None) is None unsupported_values = [None, [], "", 20, " ", 0.01, "new_value"] for value in unsupported_values: error = {ct.err_code: 1, ct.err_msg: "invalid mmap.enabled value"} self.alter_index_properties(client, collection_name, idx_names[0], properties={"mmap.enabled": value}, check_task=CheckTasks.err_res, check_items=error) @pytest.mark.tags(CaseLabel.L1) def test_milvus_client_create_index_idempotent_with_field_warmup(self): """ target: test create index idempotency when warmup is set at field level method: 1. create collection 2. release collection 3. alter field to set warmup=sync at field level (this adds warmup to field TypeParams) 4. drop existing index 5. create index (first time) 6. create same index again (second time - should be idempotent) expected: both create_index calls should succeed (idempotent behavior) issue: https://github.com/milvus-io/milvus/issues/XXXXX note: This test verifies the fix for checkParams function in index_meta.go which was missing WarmupKey in DeleteParams, causing idempotency check to fail. When index is created, WarmupKey is removed from stored TypeParams. When checking idempotency, WarmupKey should also be filtered from request TypeParams. """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. create collection self.create_collection(client, collection_name, default_dim, consistency_level="Strong") # 2. release collection before altering field properties self.release_collection(client, collection_name) # 3. alter field to set warmup=sync at field level # This adds warmup to field TypeParams, which will be included in CreateIndex request self.alter_collection_field( client, collection_name, field_name=default_vector_field_name, field_params={"warmup": "sync"} ) # 4. drop existing index self.drop_index(client, collection_name, default_vector_field_name) res = self.list_indexes(client, collection_name)[0] assert res == [] # 5. prepare index params index_params = self.prepare_index_params(client)[0] index_params.add_index( field_name=default_vector_field_name, index_type="HNSW", metric_type="L2", params={"M": 8, "efConstruction": 200}, ) # 6. create index (first time) - should succeed self.create_index(client, collection_name, index_params) idx_names, _ = self.list_indexes(client, collection_name, field_name=default_vector_field_name) assert len(idx_names) == 1 # 7. create same index again (second time) - should be idempotent and succeed # Before fix: this would fail with "at most one distinct index is allowed per field" # because checkParams didn't filter WarmupKey from TypeParams comparison self.create_index(client, collection_name, index_params) idx_names_after, _ = self.list_indexes(client, collection_name, field_name=default_vector_field_name) assert len(idx_names_after) == 1 assert idx_names == idx_names_after # 8. cleanup self.drop_collection(client, collection_name) @pytest.mark.tags(CaseLabel.L1) def test_milvus_client_create_index_idempotent_with_collection_warmup(self): """ target: test create index idempotency when warmup is set at collection level method: 1. create collection 2. release collection 3. alter collection properties to set warmup.vectorField=sync 4. drop existing index 5. create index (first time) 6. create same index again (second time - should be idempotent) expected: both create_index calls should succeed (idempotent behavior) note: This test verifies the fix for checkParams function in index_meta.go using collection-level warmup settings (warmup.vectorField) """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. create collection self.create_collection(client, collection_name, default_dim, consistency_level="Strong") # 2. release collection before altering properties self.release_collection(client, collection_name) # 3. alter collection properties to set warmup.vectorField=sync self.alter_collection_properties(client, collection_name, properties={"warmup.vectorField": "sync"}) # 4. drop existing index self.drop_index(client, collection_name, default_vector_field_name) res = self.list_indexes(client, collection_name)[0] assert res == [] # 5. prepare index params index_params = self.prepare_index_params(client)[0] index_params.add_index( field_name=default_vector_field_name, index_type="HNSW", metric_type="L2", params={"M": 8, "efConstruction": 200}, ) # 6. create index (first time) - should succeed self.create_index(client, collection_name, index_params) idx_names, _ = self.list_indexes(client, collection_name, field_name=default_vector_field_name) assert len(idx_names) == 1 # 7. create same index again (second time) - should be idempotent and succeed self.create_index(client, collection_name, index_params) idx_names_after, _ = self.list_indexes(client, collection_name, field_name=default_vector_field_name) assert len(idx_names_after) == 1 assert idx_names == idx_names_after # 8. cleanup self.drop_collection(client, collection_name) class TestMilvusClientAlterCollection(TestMilvusClientV2Base): @pytest.mark.tags(CaseLabel.L0) def test_milvus_client_alter_collection_default(self): """ target: test alter collection method: 1. alter collection properties after load verify alter successfully if trying to altering mmap.enabled or collection.ttl.seconds 2. alter collection properties after release verify alter successfully 3. drop collection properties after load verify drop successfully expected: alter successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() self.create_collection(client, collection_name, ct.default_dim, consistency_level="Strong") self.load_collection(client, collection_name) res1 = self.describe_collection(client, collection_name)[0] assert len(res1.get("properties", {})) >= 1 # 1. alter collection properties after load self.load_collection(client, collection_name) error = {ct.err_code: 999, ct.err_msg: "can not alter mmap properties if collection loaded"} self.alter_collection_properties( client, collection_name, properties={"mmap.enabled": True}, check_task=CheckTasks.err_res, check_items=error ) error = {ct.err_code: 999, ct.err_msg: "can not delete mmap properties if collection loaded"} self.drop_collection_properties( client, collection_name, property_keys=["mmap.enabled"], check_task=CheckTasks.err_res, check_items=error ) # TODO error = {ct.err_code: 999, ct.err_msg: "cannot delete key dynamicfield.enabled"} self.drop_collection_properties( client, collection_name, property_keys=["dynamicfield.enabled"], check_task=CheckTasks.err_res, check_items=error, ) self.describe_collection(client, collection_name)[0] self.drop_collection_properties(client, collection_name, property_keys=["collection.ttl.seconds"]) # 2. alter collection properties after release self.release_collection(client, collection_name) self.alter_collection_properties(client, collection_name, properties={"mmap.enabled": True}) res2 = self.describe_collection(client, collection_name)[0] assert {"mmap.enabled": "True"}.items() <= res2.get("properties", {}).items() self.alter_collection_properties(client, collection_name, properties={"collection.ttl.seconds": 100}) res2 = self.describe_collection(client, collection_name)[0] assert {"mmap.enabled": "True", "collection.ttl.seconds": "100"}.items() <= res2.get("properties", {}).items() self.drop_collection_properties( client, collection_name, property_keys=["mmap.enabled", "collection.ttl.seconds"] ) res3 = self.describe_collection(client, collection_name)[0] assert "mmap.enabled" not in res3.get("properties", {}) assert "collection.ttl.seconds" not in res3.get("properties", {}) @pytest.mark.tags(CaseLabel.L1) def test_milvus_client_alter_enable_dynamic_collection_field(self): """ target: test enable dynamic field and mixed field operations method: create collection, add field, enable dynamic field, insert mixed data, query/search expected: dynamic field works with new field and static field """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() dim = 8 # 1. create collection schema = self.create_schema(client, enable_dynamic_field=False)[0] schema.add_field(default_primary_key_field_name, DataType.INT64, max_length=64, is_primary=True, auto_id=False) schema.add_field(default_vector_field_name, DataType.FLOAT_VECTOR, dim=dim) schema.add_field(default_string_field_name, DataType.VARCHAR, max_length=64, is_partition_key=True) index_params = self.prepare_index_params(client)[0] index_params.add_index(default_vector_field_name, metric_type="COSINE") self.create_collection(client, collection_name, dimension=dim, schema=schema, index_params=index_params) # 2. Prepare and insert data schema_info = self.describe_collection(client, collection_name)[0] rows = cf.gen_row_data_by_schema(nb=default_nb, schema=schema_info) results = self.insert(client, collection_name, rows)[0] assert results["insert_count"] == default_nb # 3. add new field default_value = 100 self.add_collection_field( client, collection_name, field_name="field_new", data_type=DataType.INT64, nullable=True, default_value=default_value, ) # 4. alter collection dynamic field enable self.alter_collection_properties(client, collection_name, {"dynamicfield.enabled": True}) res = self.describe_collection(client, collection_name)[0] assert res.get("enable_dynamic_field", None) is True # 5. insert data with dynamic field and new field vectors = cf.gen_vectors(default_nb, dim, vector_data_type=DataType.FLOAT_VECTOR) rows_new = [ { default_primary_key_field_name: i, default_vector_field_name: vectors[i], default_string_field_name: str(i), default_new_field_name: i, default_dynamic_field_name: {"a": {"b": i}}, } for i in range(default_nb) ] self.insert(client, collection_name, rows_new) # 6. create index index_params = self.prepare_index_params(client)[0] index_params.add_index( field_name=default_dynamic_field_name, index_type="INVERTED", params={"json_cast_type": "DOUBLE", "json_path": f"{default_dynamic_field_name}['a']['b']"}, ) self.create_index(client, collection_name, index_params) index_name = "$meta/" + default_dynamic_field_name self.describe_index( client, collection_name, index_name + "/a/b", check_task=CheckTasks.check_describe_index_property, check_items={ "json_cast_type": "DOUBLE", "json_path": f"{default_dynamic_field_name}['a']['b']", "index_type": "INVERTED", "field_name": default_dynamic_field_name, "index_name": index_name + "/a/b", }, ) # 7. query using filter with dynamic field and new field res = self.query( client, collection_name, filter=f"{default_dynamic_field_name}['a']['b'] >= 0 and field_new < {default_value}", output_fields=[default_dynamic_field_name], check_task=CheckTasks.check_query_results, check_items={ exp_res: [ {"id": item["id"], default_dynamic_field_name: item[default_dynamic_field_name]} for item in rows_new ] }, )[0] assert set(res[0].keys()) == {default_dynamic_field_name, default_primary_key_field_name} # 8. search using filter with dynamic field and new field vectors_to_search = [vectors[0]] insert_ids = [i for i in range(default_nb)] self.search( client, collection_name, vectors_to_search, filter=f"{default_dynamic_field_name}['a']['b'] >= 0 and field_new < {default_value}", check_task=CheckTasks.check_search_results, check_items={ "enable_milvus_client_api": True, "nq": len(vectors_to_search), "ids": insert_ids, "pk_name": default_primary_key_field_name, "limit": default_limit, }, ) # 9. add new field same as dynamic field name self.add_collection_field( client, collection_name, field_name=default_dynamic_field_name, data_type=DataType.INT64, nullable=True, default_value=default_value, ) # 10. query using filter with dynamic field and new field res = self.query( client, collection_name, filter=f'$meta["{default_dynamic_field_name}"]["a"]["b"] >= 0 and {default_dynamic_field_name} == {default_value}', output_fields=[default_dynamic_field_name, f'$meta["{default_dynamic_field_name}"]'], check_task=CheckTasks.check_query_results, check_items={exp_res: [{"id": item["id"], default_dynamic_field_name: default_value} for item in rows_new]}, )[0] # dynamic field same as new field name, output_fields contain dynamic field, result do not contain dynamic field # https://github.com/milvus-io/milvus/issues/41702 assert set(res[0].keys()) == {default_dynamic_field_name, default_primary_key_field_name} # 11. search using filter with dynamic field and new field self.search( client, collection_name, vectors_to_search, filter=f'$meta["{default_dynamic_field_name}"]["a"]["b"] >= 0 and {default_dynamic_field_name} == {default_value}', check_task=CheckTasks.check_search_results, check_items={ "enable_milvus_client_api": True, "nq": len(vectors_to_search), "ids": insert_ids, "pk_name": default_primary_key_field_name, "limit": default_limit, }, ) @pytest.mark.tags(CaseLabel.L2) @pytest.mark.parametrize("old_dynamic_flag, new_dynamic_flag", [(True, True), (False, False)]) def test_milvus_client_alter_dynamic_collection_field_no_op(self, old_dynamic_flag, new_dynamic_flag): """ target: test dynamic field no-op alter operations method: create collection with dynamic flag, alter to same flag, verify unchanged expected: no-op alter succeeds without state change """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() dim = 8 # 1. create collection schema = self.create_schema(client, enable_dynamic_field=old_dynamic_flag)[0] schema.add_field(default_primary_key_field_name, DataType.INT64, max_length=64, is_primary=True, auto_id=False) schema.add_field(default_vector_field_name, DataType.FLOAT_VECTOR, dim=dim) schema.add_field(default_string_field_name, DataType.VARCHAR, max_length=64, is_partition_key=True) index_params = self.prepare_index_params(client)[0] index_params.add_index(default_vector_field_name, metric_type="COSINE") self.create_collection(client, collection_name, dimension=dim, schema=schema, index_params=index_params) # 2. alter collection dynamic field self.alter_collection_properties(client, collection_name, properties={"dynamicfield.enabled": new_dynamic_flag}) res = self.describe_collection(client, collection_name)[0] assert res.get("enable_dynamic_field", None) is new_dynamic_flag @pytest.mark.tags(CaseLabel.L1) @pytest.mark.parametrize("pk_field_type", [DataType.INT64, DataType.VARCHAR]) def test_milvus_client_alter_allow_insert_auto_id(self, pk_field_type): """ target: test alter collection allow insert auto id method: 1. create collection with auto_id=True 2. try to insert data with primary key 3. verify insert failed 4. alter collection allow_insert_auto_id=True 5. insert data with customized primary key 6. verify insert successfully 7. verify the new inserted data's primary keys are customized 8. verify the collection info 9. drop the collection properties allow_insert_auto_id 10. alter collection allow_insert_auto_id=False 11. verify the collection info 12. alter collection allow_insert_auto_id=True with string value 13. verify the collection info 14. insert data with customized primary key 15. verify insert successfully expected: insert successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() dim = 8 # 1. create collection schema = self.create_schema(client, enable_dynamic_field=False)[0] schema.add_field(default_primary_key_field_name, pk_field_type, max_length=64, is_primary=True, auto_id=True) schema.add_field(default_vector_field_name, DataType.FLOAT_VECTOR, dim=dim) index_params = self.prepare_index_params(client)[0] index_params.add_index(default_vector_field_name, metric_type="COSINE") self.create_collection(client, collection_name, dimension=dim, schema=schema, index_params=index_params) # 2. try to insert data with primary key rows_with_pk = [ { default_primary_key_field_name: i, default_vector_field_name: cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)[0], } for i in range(100) ] if pk_field_type == DataType.VARCHAR: rows_with_pk = [ { default_primary_key_field_name: f"id_{i}", default_vector_field_name: cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)[0], } for i in range(100) ] error = {ct.err_code: 999, ct.err_msg: "more fieldData has pass in"} self.insert(client, collection_name, rows_with_pk, check_task=CheckTasks.err_res, check_items=error) rows_without_pk = cf.gen_row_data_by_schema(nb=100, schema=schema) self.insert(client, collection_name, rows_without_pk) self.flush(client, collection_name) num_entities = self.get_collection_stats(client, collection_name)[0] assert num_entities.get("row_count", None) == 100 self.load_collection(client, collection_name) filter = f"{default_primary_key_field_name} in [10, 20,90]" if pk_field_type == DataType.VARCHAR: filter = f"{default_primary_key_field_name} in ['id_10', 'id_20', 'id_90']" res = self.query(client, collection_name, filter=filter, output_fields=[default_primary_key_field_name])[0] assert (len(res)) == 0 # 3. alter collection allow_insert_auto_id=True self.alter_collection_properties(client, collection_name, properties={"allow_insert_auto_id": True}) # 4. insert data with customized primary key self.insert(client, collection_name, rows_with_pk) # 5. verify insert successfully self.flush(client, collection_name) num_entities = self.get_collection_stats(client, collection_name)[0] assert num_entities.get("row_count", None) == 100 * 2 # 6. verify the new inserted data's primary keys are customized res = self.query(client, collection_name, filter=filter, output_fields=[default_primary_key_field_name])[0] assert (len(res)) == 3 # check the collection info res = self.describe_collection(client, collection_name)[0] assert res.get("properties").get("allow_insert_auto_id", None) == "True" # drop the collection properties allow_insert_auto_id self.drop_collection_properties(client, collection_name, property_keys=["allow_insert_auto_id"]) res = self.describe_collection(client, collection_name)[0] assert res.get("properties").get("allow_insert_auto_id", None) is None self.insert(client, collection_name, rows_with_pk, check_task=CheckTasks.err_res, check_items=error) # alter collection allow_insert_auto_id=False self.alter_collection_properties(client, collection_name, properties={"allow_insert_auto_id": False}) res = self.describe_collection(client, collection_name)[0] assert res.get("properties").get("allow_insert_auto_id", None) == "False" self.insert(client, collection_name, rows_with_pk, check_task=CheckTasks.err_res, check_items=error) # alter collection allow_insert_auto_id=True with string value self.alter_collection_properties(client, collection_name, properties={"allow_insert_auto_id": "True"}) res = self.describe_collection(client, collection_name)[0] assert res.get("properties").get("allow_insert_auto_id", None) == "True" rows_with_pk = [ { default_primary_key_field_name: i, default_vector_field_name: cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)[0], } for i in range(100, 200) ] if pk_field_type == DataType.VARCHAR: rows_with_pk = [ { default_primary_key_field_name: f"id_{i}", default_vector_field_name: cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)[0], } for i in range(100, 200) ] self.insert(client, collection_name, rows_with_pk) self.flush(client, collection_name) num_entities = self.get_collection_stats(client, collection_name)[0] assert num_entities.get("row_count", None) == 100 * 3 class TestMilvusClientAlterCollectionField(TestMilvusClientV2Base): @pytest.mark.tags(CaseLabel.L0) @pytest.mark.parametrize("add_field", [True, False]) def test_milvus_client_alter_collection_field_default(self, add_field): """ target: test alter collection field before load method: alter varchar field max length expected: alter successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. create collection schema = self.create_schema(client, enable_dynamic_field=False)[0] dim = 32 pk_field_name = "id_string" vector_field_name = "embeddings" str_field_name = "title" json_field_name = "json_field" array_field_name = "tags" new_field_name = "field_new" max_length = 16 schema.add_field(pk_field_name, DataType.VARCHAR, max_length=max_length, is_primary=True, auto_id=False) schema.add_field(vector_field_name, DataType.FLOAT_VECTOR, dim=dim, mmap_enabled=True) schema.add_field(str_field_name, DataType.VARCHAR, max_length=max_length, mmap_enabled=True) schema.add_field(json_field_name, DataType.JSON, mmap_enabled=False) schema.add_field( field_name=array_field_name, datatype=DataType.ARRAY, element_type=DataType.VARCHAR, max_capacity=10, max_length=max_length, ) index_params = self.prepare_index_params(client)[0] index_params.add_index( field_name=vector_field_name, metric_type="COSINE", index_type="IVF_FLAT", params={"nlist": 128} ) index_params.add_index(field_name=str_field_name) self.create_collection(client, collection_name, schema=schema, index_params=index_params) check_items = { str_field_name: {"max_length": max_length, "mmap_enabled": True}, vector_field_name: {"mmap_enabled": True}, json_field_name: {"mmap_enabled": False}, } if add_field: self.add_collection_field( client, collection_name, field_name="field_new", data_type=DataType.VARCHAR, nullable=True, max_length=max_length, ) check_items["field_new"] = {"max_length": max_length} self.describe_collection( client, collection_name, check_task=CheckTasks.check_collection_fields_properties, check_items=check_items ) rng = np.random.default_rng(seed=19530) rows = [ { pk_field_name: f"id_{i}", vector_field_name: list(rng.random((1, dim))[0]), str_field_name: cf.gen_str_by_length(max_length), json_field_name: {"number": i}, array_field_name: [cf.gen_str_by_length(max_length) for _ in range(10)], # add new field data (only when add_field is True) **({"field_new": cf.gen_str_by_length(max_length)} if add_field else {}), } for i in range(ct.default_nb) ] self.insert(client, collection_name, rows) # 1. alter collection field before load self.release_collection(client, collection_name) new_max_length = max_length // 2 # TODO: use one format of mmap_enabled after #38443 fixed self.alter_collection_field( client, collection_name, field_name=str_field_name, field_params={"max_length": new_max_length, "mmap.enabled": False}, ) self.alter_collection_field( client, collection_name, field_name=pk_field_name, field_params={"max_length": new_max_length} ) self.alter_collection_field( client, collection_name, field_name=json_field_name, field_params={"mmap.enabled": True} ) self.alter_collection_field( client, collection_name, field_name=vector_field_name, field_params={"mmap.enabled": False} ) self.alter_collection_field( client, collection_name, field_name=array_field_name, field_params={"max_length": new_max_length} ) self.alter_collection_field( client, collection_name, field_name=array_field_name, field_params={"max_capacity": 20} ) error = {ct.err_code: 999, ct.err_msg: "can not modify the maxlength for non-string types"} self.alter_collection_field( client, collection_name, field_name=vector_field_name, field_params={"max_length": new_max_length}, check_task=CheckTasks.err_res, check_items=error, ) error = {ct.err_code: 999, ct.err_msg: "element_type does not allow update in collection field param"} self.alter_collection_field( client, collection_name, field_name=array_field_name, field_params={"element_type": DataType.INT64}, check_task=CheckTasks.err_res, check_items=error, ) check_items_new = { str_field_name: {"max_length": new_max_length, "mmap_enabled": False}, vector_field_name: {"mmap_enabled": False}, json_field_name: {"mmap_enabled": True}, array_field_name: {"max_length": new_max_length, "max_capacity": 20}, } if add_field: self.alter_collection_field( client, collection_name, field_name="field_new", field_params={"max_length": new_max_length} ) check_items_new["field_new"] = {"max_length": new_max_length} self.describe_collection( client, collection_name, check_task=CheckTasks.check_collection_fields_properties, check_items=check_items_new, ) # verify that cannot insert data with the old max_length fields_to_verify = [pk_field_name, str_field_name, array_field_name] if add_field: fields_to_verify.append(new_field_name) for alter_field in fields_to_verify: error = {ct.err_code: 999, ct.err_msg: f"length of varchar field {alter_field} exceeds max length"} if alter_field == array_field_name: error = { ct.err_code: 999, ct.err_msg: f'length of Array array field "{array_field_name}" exceeds max length', } rows = [ { pk_field_name: cf.gen_str_by_length(max_length) if alter_field == pk_field_name else f"id_{i}", vector_field_name: list(rng.random((1, dim))[0]), str_field_name: cf.gen_str_by_length(max_length) if alter_field == str_field_name else f"ti_{i}", json_field_name: {"number": i}, array_field_name: [cf.gen_str_by_length(max_length) for _ in range(10)] if alter_field == array_field_name else [f"tags_{j}" for j in range(10)], **( {"field_new": cf.gen_str_by_length(max_length)} if add_field and alter_field == new_field_name else {} ), } for i in range(ct.default_nb, ct.default_nb + 10) ] self.insert(client, collection_name, rows, check_task=CheckTasks.err_res, check_items=error) # verify that can insert data with the new max_length rows = [ { pk_field_name: f"new_{cf.gen_str_by_length(new_max_length - 4)}", vector_field_name: list(rng.random((1, dim))[0]), str_field_name: cf.gen_str_by_length(new_max_length), json_field_name: {"number": i}, array_field_name: [cf.gen_str_by_length(new_max_length) for _ in range(10)], **({"field_new": cf.gen_str_by_length(new_max_length)} if add_field else {}), } for i in range(ct.default_nb, ct.default_nb + 10) ] self.insert(client, collection_name, rows) # 2. alter collection field after load self.load_collection(client, collection_name) error = {ct.err_code: 999, ct.err_msg: "can not alter collection field properties if collection loaded"} self.alter_collection_field( client, collection_name, field_name=str_field_name, field_params={"max_length": max_length, "mmap.enabled": True}, check_task=CheckTasks.err_res, check_items=error, ) self.alter_collection_field( client, collection_name, field_name=vector_field_name, field_params={"mmap.enabled": True}, check_task=CheckTasks.err_res, check_items=error, ) self.alter_collection_field( client, collection_name, field_name=pk_field_name, field_params={"max_length": max_length} ) if add_field: self.alter_collection_field( client, collection_name, field_name=new_field_name, field_params={"max_length": max_length} ) res = self.query(client, collection_name, filter=f"{pk_field_name} in ['id_10', 'id_20']", output_fields=["*"])[ 0 ] assert (len(res)) == 2 res = self.query(client, collection_name, filter=f"{pk_field_name} like 'new_%'", output_fields=["*"])[0] assert (len(res)) == 10 @pytest.mark.tags(CaseLabel.L1) def test_milvus_client_alter_collection_field_nullable_field(self): """ target: test alter collection field with nullable field method: create collection with nullable field and alter field expected: alter successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() dim = 8 # create collection schema = self.create_schema(client, enable_dynamic_field=False)[0] schema.add_field("id_string", DataType.VARCHAR, max_length=64, is_primary=True, auto_id=False) schema.add_field("embeddings_1", DataType.FLOAT_VECTOR, dim=dim, nullable=True) schema.add_field("embeddings_2", DataType.FLOAT_VECTOR, dim=dim) schema.add_field("varchar_1", DataType.VARCHAR, max_length=64, nullable=True) schema.add_field("varchar_2", DataType.VARCHAR, max_length=64) self.create_collection(client, collection_name, dimension=dim, schema=schema) # try to alert nullable vector field to non-nullable field error = {ct.err_code: 999, ct.err_msg: "nullable does not allow update in collection field param"} self.alter_collection_field( client, collection_name, field_name="embeddings_1", field_params={"nullable": False}, check_task=CheckTasks.err_res, check_items=error, ) # try to alert non-nullable vector field to nullable field self.alter_collection_field( client, collection_name, field_name="embeddings_2", field_params={"nullable": True}, check_task=CheckTasks.err_res, check_items=error, ) # try to alert nullable varchar field to non-nullable varchar field self.alter_collection_field( client, collection_name, field_name="varchar_1", field_params={"nullable": False}, check_task=CheckTasks.err_res, check_items=error, ) # try to alert non-nullable varchar field to nullable varchar field self.alter_collection_field( client, collection_name, field_name="varchar_2", field_params={"nullable": True}, check_task=CheckTasks.err_res, check_items=error, ) # add a nullable vector field to the collection self.add_collection_field( client, collection_name, field_name="embeddings_3", data_type=DataType.FLOAT_VECTOR, dim=dim, nullable=True ) # try to alert the new added nullable vector field to non-nullable field self.alter_collection_field( client, collection_name, field_name="embeddings_3", field_params={"nullable": False}, check_task=CheckTasks.err_res, check_items=error, ) # add a nullable varchar field to the collection self.add_collection_field( client, collection_name, field_name="varchar_3", data_type=DataType.VARCHAR, max_length=64, nullable=True ) # try to alert the new added nullable varchar field to non-nullable varchar field self.alter_collection_field( client, collection_name, field_name="varchar_3", field_params={"nullable": False}, check_task=CheckTasks.err_res, check_items=error, ) # drop the collection self.drop_collection(client, collection_name) class TestMilvusClientAlterDatabase(TestMilvusClientV2Base): @pytest.mark.tags(CaseLabel.L0) def test_milvus_client_alter_database_default(self): """ target: test alter database method: 1. alter database properties before load alter successfully 2. alter database properties after load alter successfully expected: alter successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() self.create_collection(client, collection_name, ct.default_dim, consistency_level="Strong") self.release_collection(client, collection_name) default_db = "default" res1 = self.describe_database(client, db_name=default_db)[0] if len(res1.keys()) != 1: self.drop_database_properties(client, db_name=default_db, property_keys=res1.keys()) assert len(self.describe_database(client, default_db)[0].keys()) == 1 for need_load in [True, False]: if need_load: log.debug("alter database after load collection") self.load_collection(client, collection_name) # 1. alter default database properties before load properties = { "key1": 1, "key2": "value2", "key3": [1, 2, 3], } self.alter_database_properties(client, db_name=default_db, properties=properties) res1 = self.describe_database(client, db_name=default_db)[0] # assert res1.properties.items() >= properties.items() assert len(res1.keys()) == 4 my_db = cf.gen_unique_str(prefix) self.create_database(client, my_db, properties=properties) res1 = self.describe_database(client, db_name=my_db)[0] # assert res1.properties.items() >= properties.items() assert len(res1.keys()) == 4 properties = {"key1": 2, "key2": "value3", "key3": [1, 2, 3], "key4": 0.123} self.alter_database_properties(client, db_name=my_db, properties=properties) res1 = self.describe_database(client, db_name=my_db)[0] # assert res1.properties.items() >= properties.items() assert len(res1.keys()) == 5 # drop the default database properties self.drop_database_properties(client, db_name=default_db, property_keys=["key1", "key2"]) res1 = self.describe_database(client, db_name=default_db)[0] assert len(res1.keys()) == 2 self.drop_database_properties(client, db_name=default_db, property_keys=["key3", "key_non_exist"]) res1 = self.describe_database(client, db_name=default_db)[0] assert len(res1.keys()) == 1 # drop the user database self.drop_database(client, my_db) class TestMilvusClientAlterCollectionFieldDescriptionValid(TestMilvusClientV2Base): """ Positive tests for alter_collection_field() to change field description PR: https://github.com/milvus-io/milvus/pull/47057 Issue: https://github.com/milvus-io/milvus/issues/46896 """ @pytest.mark.tags(CaseLabel.L0) def test_alter_field_description_basic(self): """ target: test basic functionality of altering field description method: create collection, alter field description, verify change expected: field description updated successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. Create collection with default schema self.create_collection(client, collection_name, default_dim) # 2. Alter field description new_description = "This is a new description for vector field" self.alter_collection_field( client, collection_name, field_name=default_vector_field_name, field_params={"field.description": new_description}, ) # 3. Verify description changed desc_res = self.describe_collection(client, collection_name)[0] for field in desc_res.get("fields", []): if field.get("name") == default_vector_field_name: assert field.get("description") == new_description, ( f"Expected description '{new_description}', got '{field.get('description')}'" ) break # 4. Cleanup self.drop_collection(client, collection_name) @pytest.mark.tags(CaseLabel.L1) def test_alter_primary_key_field_description(self): """ target: test altering primary key field description method: alter the description of primary key field expected: description updated successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. Create collection self.create_collection(client, collection_name, default_dim) # 2. Alter primary key field description new_description = "Primary key field for entity identification" self.alter_collection_field( client, collection_name, field_name=default_primary_key_field_name, field_params={"field.description": new_description}, ) # 3. Verify desc_res = self.describe_collection(client, collection_name)[0] for field in desc_res.get("fields", []): if field.get("name") == default_primary_key_field_name: assert field.get("description") == new_description break # 4. Cleanup self.drop_collection(client, collection_name) @pytest.mark.tags(CaseLabel.L1) def test_alter_vector_field_description(self): """ target: test altering vector field description method: alter the description of vector field expected: description updated successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. Create collection with custom schema schema = self.create_schema(client, enable_dynamic_field=True)[0] schema.add_field("pk", DataType.INT64, is_primary=True, auto_id=False) schema.add_field("embedding", DataType.FLOAT_VECTOR, dim=default_dim, description="original description") self.create_collection(client, collection_name, schema=schema) # 2. Alter vector field description new_description = "Dense embedding vector for similarity search" self.alter_collection_field( client, collection_name, field_name="embedding", field_params={"field.description": new_description} ) # 3. Verify desc_res = self.describe_collection(client, collection_name)[0] for field in desc_res.get("fields", []): if field.get("name") == "embedding": assert field.get("description") == new_description break # 4. Cleanup self.drop_collection(client, collection_name) @pytest.mark.tags(CaseLabel.L1) def test_alter_scalar_field_description(self): """ target: test altering scalar field description method: alter the description of a scalar field (VarChar) expected: description updated successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. Create collection with scalar field schema = self.create_schema(client, enable_dynamic_field=True)[0] schema.add_field("pk", DataType.INT64, is_primary=True, auto_id=False) schema.add_field(default_vector_field_name, DataType.FLOAT_VECTOR, dim=default_dim) schema.add_field("title", DataType.VARCHAR, max_length=256, description="original title description") self.create_collection(client, collection_name, schema=schema) # 2. Alter scalar field description new_description = "Title of the document" self.alter_collection_field( client, collection_name, field_name="title", field_params={"field.description": new_description} ) # 3. Verify desc_res = self.describe_collection(client, collection_name)[0] for field in desc_res.get("fields", []): if field.get("name") == "title": assert field.get("description") == new_description break # 4. Cleanup self.drop_collection(client, collection_name) @pytest.mark.tags(CaseLabel.L1) def test_alter_field_description_multiple_times(self): """ target: test altering field description multiple times method: alter the same field description consecutively expected: each alteration succeeds, idempotent behavior """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. Create collection self.create_collection(client, collection_name, default_dim) # 2. Alter description multiple times descriptions = ["First description", "Second description", "Third description", "Final description"] for desc in descriptions: self.alter_collection_field( client, collection_name, field_name=default_vector_field_name, field_params={"field.description": desc} ) # Verify each change desc_res = self.describe_collection(client, collection_name)[0] for field in desc_res.get("fields", []): if field.get("name") == default_vector_field_name: assert field.get("description") == desc break # 3. Cleanup self.drop_collection(client, collection_name) @pytest.mark.tags(CaseLabel.L2) def test_alter_field_description_to_empty(self): """ target: test altering field description to empty string method: set field description to empty string expected: description cleared successfully """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. Create collection with field having description schema = self.create_schema(client, enable_dynamic_field=True)[0] schema.add_field("pk", DataType.INT64, is_primary=True, auto_id=False) schema.add_field( default_vector_field_name, DataType.FLOAT_VECTOR, dim=default_dim, description="This field has a description", ) self.create_collection(client, collection_name, schema=schema) # 2. Clear description self.alter_collection_field( client, collection_name, field_name=default_vector_field_name, field_params={"field.description": ""} ) # 3. Verify description is empty desc_res = self.describe_collection(client, collection_name)[0] for field in desc_res.get("fields", []): if field.get("name") == default_vector_field_name: assert field.get("description") == "" break # 4. Cleanup self.drop_collection(client, collection_name) @pytest.mark.tags(CaseLabel.L2) @pytest.mark.parametrize( "description", [ "中文描述测试", "Description with emoji 🚀🎉", "Line1\nLine2\nLine3", "Tab\tseparated\tvalues", "Special chars: !@#$%^&*()_+-=[]{}|;':\",./<>?", " " * 100, # Many spaces ], ) def test_alter_field_description_special_characters(self, description): """ target: test altering field description with special characters method: set description containing unicode, emoji, newlines, etc. expected: description updated successfully with special characters preserved """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. Create collection self.create_collection(client, collection_name, default_dim) # 2. Alter with special description self.alter_collection_field( client, collection_name, field_name=default_vector_field_name, field_params={"field.description": description}, ) # 3. Verify desc_res = self.describe_collection(client, collection_name)[0] for field in desc_res.get("fields", []): if field.get("name") == default_vector_field_name: assert field.get("description") == description break # 4. Cleanup self.drop_collection(client, collection_name) class TestMilvusClientAlterCollectionFieldDescriptionInvalid(TestMilvusClientV2Base): """ Negative tests for alter_collection_field() to change field description PR: https://github.com/milvus-io/milvus/pull/47057 """ @pytest.mark.tags(CaseLabel.L1) def test_alter_field_description_nonexistent_collection(self): """ target: test altering field description on non-existent collection method: call alter_collection_field with non-existent collection name expected: raise exception with error code 100 """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # Do not create collection, directly alter error = {ct.err_code: 100, ct.err_msg: "collection not found"} self.alter_collection_field( client, collection_name, field_name=default_vector_field_name, field_params={"field.description": "new description"}, check_task=CheckTasks.err_res, check_items=error, ) @pytest.mark.tags(CaseLabel.L1) def test_alter_field_description_nonexistent_field(self): """ target: test altering description of non-existent field method: call alter_collection_field with non-existent field name expected: raise exception indicating field not found """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. Create collection self.create_collection(client, collection_name, default_dim) # 2. Alter non-existent field error = {ct.err_code: 1100, ct.err_msg: "does not exist in collection"} self.alter_collection_field( client, collection_name, field_name="nonexistent_field", field_params={"field.description": "new description"}, check_task=CheckTasks.err_res, check_items=error, ) # 3. Cleanup self.drop_collection(client, collection_name) @pytest.mark.tags(CaseLabel.L1) def test_alter_field_description_empty_collection_name(self): """ target: test altering field description with empty collection name method: call alter_collection_field with empty string collection name expected: raise exception with error code 1 """ client = self._client() collection_name = "" error = {ct.err_code: 1, ct.err_msg: f"`collection_name` value {collection_name} is illegal"} self.alter_collection_field( client, collection_name, field_name=default_vector_field_name, field_params={"field.description": "new description"}, check_task=CheckTasks.err_res, check_items=error, ) @pytest.mark.tags(CaseLabel.L1) def test_alter_field_description_empty_field_name(self): """ target: test altering field description with empty field name method: call alter_collection_field with empty string field name expected: raise exception indicating invalid field name """ client = self._client() collection_name = cf.gen_collection_name_by_testcase_name() # 1. Create collection self.create_collection(client, collection_name, default_dim) # 2. Alter with empty field name error = {ct.err_code: 1100, ct.err_msg: "does not exist in collection"} self.alter_collection_field( client, collection_name, field_name="", field_params={"field.description": "new description"}, check_task=CheckTasks.err_res, check_items=error, ) # 3. Cleanup self.drop_collection(client, collection_name)