mirror of
https://github.com/milvus-io/milvus.git
synced 2026-07-21 10:15:43 +00:00
test: Add comprehensive test coverage for nullable vector field functionality (#47847)
## Summary Add 6 new test cases to `test_milvus_client_collection.py` covering gaps in nullable vector field test coverage, identified after verifying fixes for #47065 and #47197. **Related issue**: #47846 ### New Test Cases | Test | Level | Coverage | |------|-------|----------| | `test_milvus_client_search_all_null_vectors` | L1 | Search on 100% null vectors (sealed + growing) | | `test_milvus_client_search_iterator_null_vector` | L1 | Search iterator with all-null and mixed data | | `test_milvus_client_query_iterator_null_vector` | L1 | Query iterator returns all rows including null vectors | | `test_milvus_client_upsert_null_vector_transitions` | L1 | Non-null↔null transitions via upsert | | `test_milvus_client_multi_vector_output_null` | L1 | Multi-vector output with nullable field | | `test_milvus_client_compact_with_null_vector` | L2 | Data integrity after compaction with null vectors | ### Test Plan - [x] All 6 tests passed against Milvus 2.6.6 (10.104.17.170) - [x] No regressions in existing null vector tests 🤖 Generated with [Claude Code](https://claude.com/claude-code) Signed-off-by: yanliang567 <82361606+yanliang567@users.noreply.github.com>
This commit is contained in:
yanliang567
@@ -972,6 +972,377 @@ class TestMilvusClientCollectionValid(TestMilvusClientV2Base):
|
||||
self.release_collection(client, collection_name)
|
||||
self.drop_collection(client, collection_name)
|
||||
|
||||
@pytest.mark.tags(CaseLabel.L1)
|
||||
def test_milvus_client_search_all_null_vectors(self):
|
||||
"""
|
||||
target: test search when all vectors are null
|
||||
method: create collection with nullable vector, insert all-null vectors, search
|
||||
expected: search returns empty results without panic
|
||||
"""
|
||||
client = self._client()
|
||||
collection_name = cf.gen_collection_name_by_testcase_name()
|
||||
dim = 128
|
||||
schema = self.create_schema(client, enable_dynamic_field=False)[0]
|
||||
schema.add_field("id", DataType.INT64, is_primary=True, auto_id=False)
|
||||
schema.add_field("vec", DataType.FLOAT_VECTOR, dim=dim, nullable=True)
|
||||
schema.add_field("text", DataType.VARCHAR, max_length=64)
|
||||
index_params = self.prepare_index_params(client)[0]
|
||||
index_params.add_index("vec", index_type="HNSW", metric_type="COSINE")
|
||||
self.create_collection(client, collection_name, schema=schema, index_params=index_params)
|
||||
|
||||
# insert 100 rows with ALL null vectors
|
||||
rows = [{"id": i, "vec": None, "text": f"text_{i}"} for i in range(100)]
|
||||
self.insert(client, collection_name, rows)
|
||||
self.flush(client, collection_name)
|
||||
|
||||
# search should return empty results (not panic)
|
||||
vectors_to_search = cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)
|
||||
self.search(client, collection_name, vectors_to_search,
|
||||
search_params={}, anns_field="vec", limit=10,
|
||||
check_task=CheckTasks.check_search_results,
|
||||
check_items={"enable_milvus_client_api": True,
|
||||
"nq": 1, "limit": 0, "pk_name": "id"})
|
||||
|
||||
# also verify on growing segment (no flush)
|
||||
rows2 = [{"id": i + 100, "vec": None, "text": f"text_{i + 100}"} for i in range(50)]
|
||||
self.insert(client, collection_name, rows2)
|
||||
self.search(client, collection_name, vectors_to_search,
|
||||
search_params={}, anns_field="vec", limit=10,
|
||||
consistency_level="Strong",
|
||||
check_task=CheckTasks.check_search_results,
|
||||
check_items={"enable_milvus_client_api": True,
|
||||
"nq": 1, "limit": 0, "pk_name": "id"})
|
||||
|
||||
self.release_collection(client, collection_name)
|
||||
self.drop_collection(client, collection_name)
|
||||
|
||||
@pytest.mark.tags(CaseLabel.L1)
|
||||
def test_milvus_client_search_iterator_null_vector(self):
|
||||
"""
|
||||
target: test search iterator with nullable vector field
|
||||
method: create collection with nullable vector, insert mixed data, run search iterator
|
||||
expected: 1. all-null: iterator returns 0 results without panic
|
||||
2. mixed: iterator only returns non-null vector rows
|
||||
"""
|
||||
client = self._client()
|
||||
collection_name = cf.gen_collection_name_by_testcase_name()
|
||||
dim = 128
|
||||
schema = self.create_schema(client, enable_dynamic_field=False)[0]
|
||||
schema.add_field("id", DataType.INT64, is_primary=True, auto_id=False)
|
||||
schema.add_field("vec", DataType.FLOAT_VECTOR, dim=dim, nullable=True)
|
||||
index_params = self.prepare_index_params(client)[0]
|
||||
index_params.add_index("vec", index_type="HNSW", metric_type="COSINE")
|
||||
self.create_collection(client, collection_name, schema=schema, index_params=index_params)
|
||||
|
||||
# insert all-null vectors
|
||||
rows = [{"id": i, "vec": None} for i in range(100)]
|
||||
self.insert(client, collection_name, rows)
|
||||
self.flush(client, collection_name)
|
||||
|
||||
vectors_to_search = cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)
|
||||
# search iterator on all-null data should return 0 results
|
||||
self.search_iterator(client, collection_name, vectors_to_search,
|
||||
batch_size=10, limit=50, anns_field="vec",
|
||||
search_params={},
|
||||
check_task=CheckTasks.check_search_iterator,
|
||||
check_items={"batch_size": 10, "iterate_times": 1})
|
||||
|
||||
# insert 50 non-null vectors (id 100-149)
|
||||
non_null_rows = [{"id": i + 100,
|
||||
"vec": cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)[0]}
|
||||
for i in range(50)]
|
||||
self.insert(client, collection_name, non_null_rows)
|
||||
self.flush(client, collection_name)
|
||||
|
||||
# search iterator on mixed data should only return non-null vector rows
|
||||
self.search_iterator(client, collection_name, vectors_to_search,
|
||||
batch_size=10, limit=50, anns_field="vec",
|
||||
search_params={},
|
||||
check_task=CheckTasks.check_search_iterator,
|
||||
check_items={"batch_size": 10})
|
||||
|
||||
self.release_collection(client, collection_name)
|
||||
self.drop_collection(client, collection_name)
|
||||
|
||||
@pytest.mark.tags(CaseLabel.L1)
|
||||
def test_milvus_client_query_iterator_null_vector(self):
|
||||
"""
|
||||
target: test query iterator with nullable vector field
|
||||
method: create collection with nullable vector, insert mixed data, run query iterator
|
||||
expected: query iterator returns all rows including those with null vectors
|
||||
"""
|
||||
client = self._client()
|
||||
collection_name = cf.gen_collection_name_by_testcase_name()
|
||||
dim = 128
|
||||
schema = self.create_schema(client, enable_dynamic_field=False)[0]
|
||||
schema.add_field("id", DataType.INT64, is_primary=True, auto_id=False)
|
||||
schema.add_field("vec", DataType.FLOAT_VECTOR, dim=dim, nullable=True)
|
||||
schema.add_field("tag", DataType.VARCHAR, max_length=64)
|
||||
index_params = self.prepare_index_params(client)[0]
|
||||
index_params.add_index("vec", index_type="HNSW", metric_type="COSINE")
|
||||
self.create_collection(client, collection_name, schema=schema, index_params=index_params)
|
||||
|
||||
# insert 100 rows: even ids have null vectors, odd ids have valid vectors
|
||||
total = 100
|
||||
rows = []
|
||||
for i in range(total):
|
||||
if i % 2 == 0:
|
||||
rows.append({"id": i, "vec": None, "tag": "null_vec"})
|
||||
else:
|
||||
rows.append({"id": i,
|
||||
"vec": cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)[0],
|
||||
"tag": "valid_vec"})
|
||||
self.insert(client, collection_name, rows)
|
||||
self.flush(client, collection_name)
|
||||
|
||||
# query iterator should return ALL rows (including null vectors)
|
||||
it = self.query_iterator(client, collection_name, batch_size=20,
|
||||
output_fields=["id", "vec", "tag"])[0]
|
||||
all_results = []
|
||||
while True:
|
||||
batch = it.next()
|
||||
if not batch:
|
||||
break
|
||||
all_results.extend(batch)
|
||||
it.close()
|
||||
assert len(all_results) == total, f"Expected {total} results but got {len(all_results)}"
|
||||
# verify null vector rows are included
|
||||
null_vec_ids = [r["id"] for r in all_results if r.get("vec") is None]
|
||||
valid_vec_ids = [r["id"] for r in all_results if r.get("vec") is not None]
|
||||
assert len(null_vec_ids) == total // 2
|
||||
assert len(valid_vec_ids) == total // 2
|
||||
for _id in null_vec_ids:
|
||||
assert _id % 2 == 0
|
||||
for _id in valid_vec_ids:
|
||||
assert _id % 2 == 1
|
||||
|
||||
self.release_collection(client, collection_name)
|
||||
self.drop_collection(client, collection_name)
|
||||
|
||||
@pytest.mark.tags(CaseLabel.L1)
|
||||
def test_milvus_client_upsert_null_vector_transitions(self):
|
||||
"""
|
||||
target: test upsert transitions between null and non-null vectors
|
||||
method: 1. insert data with valid vectors
|
||||
2. upsert same PKs with null vectors -> verify search cannot find them
|
||||
3. upsert same PKs back with valid vectors -> verify search finds them
|
||||
expected: upsert correctly transitions vectors between null and non-null states
|
||||
"""
|
||||
client = self._client()
|
||||
collection_name = cf.gen_collection_name_by_testcase_name()
|
||||
dim = 128
|
||||
schema = self.create_schema(client, enable_dynamic_field=False)[0]
|
||||
schema.add_field("id", DataType.INT64, is_primary=True, auto_id=False)
|
||||
schema.add_field("vec", DataType.FLOAT_VECTOR, dim=dim, nullable=True)
|
||||
schema.add_field("tag", DataType.VARCHAR, max_length=64)
|
||||
index_params = self.prepare_index_params(client)[0]
|
||||
index_params.add_index("vec", index_type="HNSW", metric_type="COSINE")
|
||||
self.create_collection(client, collection_name, schema=schema, index_params=index_params)
|
||||
|
||||
# step 1: insert 20 rows with valid vectors
|
||||
nb = 20
|
||||
vectors = cf.gen_vectors(nb, dim, vector_data_type=DataType.FLOAT_VECTOR)
|
||||
rows = [{"id": i, "vec": vectors[i], "tag": "valid"} for i in range(nb)]
|
||||
self.insert(client, collection_name, rows)
|
||||
self.flush(client, collection_name)
|
||||
|
||||
# verify all 20 rows can be found by search
|
||||
vectors_to_search = cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)
|
||||
self.search(client, collection_name, vectors_to_search,
|
||||
search_params={}, anns_field="vec", limit=nb,
|
||||
consistency_level="Strong",
|
||||
check_task=CheckTasks.check_search_results,
|
||||
check_items={"enable_milvus_client_api": True,
|
||||
"nq": 1, "limit": nb, "pk_name": "id"})
|
||||
|
||||
# step 2: upsert ids 0-9 with null vectors (non-null -> null)
|
||||
upsert_null_rows = [{"id": i, "vec": None, "tag": "null"} for i in range(10)]
|
||||
self.upsert(client, collection_name, upsert_null_rows)
|
||||
self.flush(client, collection_name)
|
||||
|
||||
# search should only find 10 rows (ids 10-19)
|
||||
self.search(client, collection_name, vectors_to_search,
|
||||
search_params={}, anns_field="vec", limit=nb,
|
||||
consistency_level="Strong",
|
||||
check_task=CheckTasks.check_search_results,
|
||||
check_items={"enable_milvus_client_api": True,
|
||||
"nq": 1, "limit": 10, "pk_name": "id"})
|
||||
# query to verify the upserted rows have tag="null"
|
||||
res = self.query(client, collection_name, filter="id < 10",
|
||||
output_fields=["id", "vec", "tag"],
|
||||
consistency_level="Strong")[0]
|
||||
for r in res:
|
||||
assert r["tag"] == "null"
|
||||
assert r["vec"] is None
|
||||
|
||||
# step 3: upsert ids 0-9 back with valid vectors (null -> non-null)
|
||||
new_vectors = cf.gen_vectors(10, dim, vector_data_type=DataType.FLOAT_VECTOR)
|
||||
upsert_valid_rows = [{"id": i, "vec": new_vectors[i], "tag": "restored"} for i in range(10)]
|
||||
self.upsert(client, collection_name, upsert_valid_rows)
|
||||
self.flush(client, collection_name)
|
||||
|
||||
# search should find all 20 rows again
|
||||
self.search(client, collection_name, vectors_to_search,
|
||||
search_params={}, anns_field="vec", limit=nb,
|
||||
consistency_level="Strong",
|
||||
check_task=CheckTasks.check_search_results,
|
||||
check_items={"enable_milvus_client_api": True,
|
||||
"nq": 1, "limit": nb, "pk_name": "id"})
|
||||
# query to verify the restored rows
|
||||
res = self.query(client, collection_name, filter="id < 10",
|
||||
output_fields=["id", "vec", "tag"],
|
||||
consistency_level="Strong")[0]
|
||||
for r in res:
|
||||
assert r["tag"] == "restored"
|
||||
assert r["vec"] is not None
|
||||
|
||||
self.release_collection(client, collection_name)
|
||||
self.drop_collection(client, collection_name)
|
||||
|
||||
@pytest.mark.tags(CaseLabel.L1)
|
||||
def test_milvus_client_multi_vector_output_null(self):
|
||||
"""
|
||||
target: test search and query output with multiple vector fields containing null vectors
|
||||
method: create collection with 2 vector fields (one nullable), insert mixed data,
|
||||
verify search and query output correctly include null vectors
|
||||
expected: output fields correctly return None for null vectors in multi-vector schema
|
||||
"""
|
||||
client = self._client()
|
||||
collection_name = cf.gen_collection_name_by_testcase_name()
|
||||
dim = 128
|
||||
schema = self.create_schema(client, enable_dynamic_field=False)[0]
|
||||
schema.add_field("id", DataType.INT64, is_primary=True, auto_id=False)
|
||||
schema.add_field("vec_required", DataType.FLOAT_VECTOR, dim=dim)
|
||||
schema.add_field("vec_nullable", DataType.FLOAT_VECTOR, dim=dim, nullable=True)
|
||||
schema.add_field("tag", DataType.VARCHAR, max_length=64)
|
||||
index_params = self.prepare_index_params(client)[0]
|
||||
index_params.add_index("vec_required", index_type="HNSW", metric_type="COSINE")
|
||||
index_params.add_index("vec_nullable", index_type="HNSW", metric_type="COSINE")
|
||||
self.create_collection(client, collection_name, schema=schema, index_params=index_params)
|
||||
|
||||
# insert 20 rows: even ids have null vec_nullable, odd ids have valid vec_nullable
|
||||
nb = 20
|
||||
rows = []
|
||||
for i in range(nb):
|
||||
row = {
|
||||
"id": i,
|
||||
"vec_required": cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)[0],
|
||||
"vec_nullable": None if i % 2 == 0 else cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)[0],
|
||||
"tag": "null" if i % 2 == 0 else "valid"
|
||||
}
|
||||
rows.append(row)
|
||||
self.insert(client, collection_name, rows)
|
||||
self.flush(client, collection_name)
|
||||
|
||||
# search on vec_required with output of both vector fields
|
||||
vectors_to_search = cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)
|
||||
res = self.search(client, collection_name, vectors_to_search,
|
||||
search_params={}, anns_field="vec_required", limit=nb,
|
||||
output_fields=["vec_required", "vec_nullable", "tag"])[0]
|
||||
assert len(res[0]) == nb
|
||||
for hit in res[0]:
|
||||
entity = hit.fields if hasattr(hit, 'fields') else hit.entity.fields
|
||||
_id = hit.id if hasattr(hit, 'id') else hit['id']
|
||||
if _id % 2 == 0:
|
||||
assert entity["vec_nullable"] is None
|
||||
assert entity["tag"] == "null"
|
||||
else:
|
||||
assert entity["vec_nullable"] is not None
|
||||
assert entity["tag"] == "valid"
|
||||
# vec_required should always be present
|
||||
assert entity["vec_required"] is not None
|
||||
|
||||
# query output with both vector fields
|
||||
res = self.query(client, collection_name, filter="id < 10",
|
||||
output_fields=["id", "vec_required", "vec_nullable", "tag"])[0]
|
||||
assert len(res) == 10
|
||||
for r in res:
|
||||
if r["id"] % 2 == 0:
|
||||
assert r["vec_nullable"] is None
|
||||
else:
|
||||
assert r["vec_nullable"] is not None
|
||||
assert r["vec_required"] is not None
|
||||
|
||||
self.release_collection(client, collection_name)
|
||||
self.drop_collection(client, collection_name)
|
||||
|
||||
@pytest.mark.tags(CaseLabel.L2)
|
||||
def test_milvus_client_compact_with_null_vector(self):
|
||||
"""
|
||||
target: test compaction with nullable vector field
|
||||
method: create collection with nullable vector, insert multiple batches with null vectors,
|
||||
trigger compaction, verify data integrity after compaction
|
||||
expected: compaction completes successfully and all data (including null vectors) is preserved
|
||||
"""
|
||||
client = self._client()
|
||||
collection_name = cf.gen_collection_name_by_testcase_name()
|
||||
dim = 128
|
||||
schema = self.create_schema(client, enable_dynamic_field=False)[0]
|
||||
schema.add_field("id", DataType.INT64, is_primary=True, auto_id=False)
|
||||
schema.add_field("vec", DataType.FLOAT_VECTOR, dim=dim, nullable=True)
|
||||
schema.add_field("tag", DataType.VARCHAR, max_length=64, is_clustering_key=True)
|
||||
index_params = self.prepare_index_params(client)[0]
|
||||
index_params.add_index("vec", index_type="HNSW", metric_type="COSINE")
|
||||
self.create_collection(client, collection_name, schema=schema, index_params=index_params)
|
||||
|
||||
# insert multiple small batches to create multiple segments for compaction
|
||||
total = 0
|
||||
null_count = 0
|
||||
valid_count = 0
|
||||
for batch in range(5):
|
||||
rows = []
|
||||
for i in range(200):
|
||||
pk = batch * 200 + i
|
||||
if pk % 3 == 0:
|
||||
rows.append({"id": pk, "vec": None, "tag": f"batch_{batch}"})
|
||||
null_count += 1
|
||||
else:
|
||||
rows.append({"id": pk,
|
||||
"vec": cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)[0],
|
||||
"tag": f"batch_{batch}"})
|
||||
valid_count += 1
|
||||
total += 1
|
||||
self.insert(client, collection_name, rows)
|
||||
self.flush(client, collection_name)
|
||||
|
||||
# verify data before compaction
|
||||
res = self.query(client, collection_name, filter="", output_fields=["count(*)"])[0]
|
||||
assert res[0]["count(*)"] == total
|
||||
|
||||
# trigger compaction
|
||||
compact_id = self.compact(client, collection_name, is_clustering=False)[0]
|
||||
start = time.time()
|
||||
while True:
|
||||
time.sleep(1)
|
||||
state = self.get_compaction_state(client, compact_id, is_clustering=False)[0]
|
||||
if state == "Completed":
|
||||
break
|
||||
if time.time() - start > 180:
|
||||
raise Exception("Compaction did not complete within 180s")
|
||||
|
||||
# verify data integrity after compaction
|
||||
res = self.query(client, collection_name, filter="", output_fields=["count(*)"])[0]
|
||||
assert res[0]["count(*)"] == total
|
||||
|
||||
# verify null vectors are still null, valid vectors are still valid
|
||||
res = self.query(client, collection_name, filter="id >= 0",
|
||||
output_fields=["id", "vec"], limit=total)[0]
|
||||
actual_null = sum(1 for r in res if r["vec"] is None)
|
||||
actual_valid = sum(1 for r in res if r["vec"] is not None)
|
||||
assert actual_null == null_count, f"Expected {null_count} null vectors, got {actual_null}"
|
||||
assert actual_valid == valid_count, f"Expected {valid_count} valid vectors, got {actual_valid}"
|
||||
|
||||
# search should still work correctly after compaction
|
||||
vectors_to_search = cf.gen_vectors(1, dim, vector_data_type=DataType.FLOAT_VECTOR)
|
||||
self.search(client, collection_name, vectors_to_search,
|
||||
search_params={}, anns_field="vec", limit=default_limit,
|
||||
check_task=CheckTasks.check_search_results,
|
||||
check_items={"enable_milvus_client_api": True,
|
||||
"nq": 1, "limit": default_limit, "pk_name": "id"})
|
||||
|
||||
self.release_collection(client, collection_name)
|
||||
self.drop_collection(client, collection_name)
|
||||
|
||||
@pytest.mark.tags(CaseLabel.L2)
|
||||
@pytest.mark.parametrize("primary_key_type", ["int64", "varchar"])
|
||||
def test_milvus_client_collection_max_fields_and_max_vector_fields(self, primary_key_type):
|
||||
|
||||
Reference in New Issue
Block a user