Collection management (#520)

* Tech spec

* Refactored Cassanda knowledge graph for single table

* Collection management, librarian services to manage metadata and collection deletion
This commit is contained in:
cybermaggedon 2025-09-18 15:57:52 +01:00 committed by GitHub
parent 48016d8fb2
commit 13ff7d765d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
48 changed files with 2941 additions and 425 deletions

View file

@ -13,163 +13,146 @@ class TestMilvusCollectionNaming:
"""Test basic collection name creation"""
result = make_safe_collection_name(
user="test_user",
collection="test_collection",
dimension=384,
collection="test_collection",
prefix="doc"
)
assert result == "doc_test_user_test_collection_384"
assert result == "doc_test_user_test_collection"
def test_make_safe_collection_name_with_special_characters(self):
"""Test collection name creation with special characters that need sanitization"""
result = make_safe_collection_name(
user="user@domain.com",
collection="test-collection.v2",
dimension=768,
prefix="entity"
)
assert result == "entity_user_domain_com_test_collection_v2_768"
assert result == "entity_user_domain_com_test_collection_v2"
def test_make_safe_collection_name_with_unicode(self):
"""Test collection name creation with Unicode characters"""
result = make_safe_collection_name(
user="测试用户",
collection="colección_española",
dimension=512,
collection="colección_española",
prefix="doc"
)
assert result == "doc_default_colecci_n_espa_ola_512"
assert result == "doc_default_colecci_n_espa_ola"
def test_make_safe_collection_name_with_spaces(self):
"""Test collection name creation with spaces"""
result = make_safe_collection_name(
user="test user",
collection="my test collection",
dimension=256,
prefix="entity"
)
assert result == "entity_test_user_my_test_collection_256"
assert result == "entity_test_user_my_test_collection"
def test_make_safe_collection_name_with_multiple_consecutive_special_chars(self):
"""Test collection name creation with multiple consecutive special characters"""
result = make_safe_collection_name(
user="user@@@domain!!!",
collection="test---collection...v2",
dimension=384,
prefix="doc"
prefix="doc"
)
assert result == "doc_user_domain_test_collection_v2_384"
assert result == "doc_user_domain_test_collection_v2"
def test_make_safe_collection_name_with_leading_trailing_underscores(self):
"""Test collection name creation with leading/trailing special characters"""
result = make_safe_collection_name(
user="__test_user__",
collection="@@test_collection##",
dimension=128,
prefix="entity"
)
assert result == "entity_test_user_test_collection_128"
assert result == "entity_test_user_test_collection"
def test_make_safe_collection_name_empty_user(self):
"""Test collection name creation with empty user (should fallback to 'default')"""
result = make_safe_collection_name(
user="",
collection="test_collection",
dimension=384,
prefix="doc"
)
assert result == "doc_default_test_collection_384"
assert result == "doc_default_test_collection"
def test_make_safe_collection_name_empty_collection(self):
"""Test collection name creation with empty collection (should fallback to 'default')"""
result = make_safe_collection_name(
user="test_user",
collection="",
dimension=384,
prefix="doc"
)
assert result == "doc_test_user_default_384"
assert result == "doc_test_user_default"
def test_make_safe_collection_name_both_empty(self):
"""Test collection name creation with both user and collection empty"""
result = make_safe_collection_name(
user="",
collection="",
dimension=384,
prefix="doc"
)
assert result == "doc_default_default_384"
assert result == "doc_default_default"
def test_make_safe_collection_name_only_special_characters(self):
"""Test collection name creation with only special characters (should fallback to 'default')"""
result = make_safe_collection_name(
user="@@@!!!",
collection="---###",
dimension=512,
prefix="entity"
)
assert result == "entity_default_default_512"
assert result == "entity_default_default"
def test_make_safe_collection_name_whitespace_only(self):
"""Test collection name creation with whitespace-only strings"""
result = make_safe_collection_name(
user=" \n\t ",
collection=" \r\n ",
dimension=256,
prefix="doc"
)
assert result == "doc_default_default_256"
assert result == "doc_default_default"
def test_make_safe_collection_name_mixed_valid_invalid_chars(self):
"""Test collection name creation with mixed valid and invalid characters"""
result = make_safe_collection_name(
user="user123@test",
collection="coll_2023.v1",
dimension=384,
prefix="entity"
)
assert result == "entity_user123_test_coll_2023_v1_384"
assert result == "entity_user123_test_coll_2023_v1"
def test_make_safe_collection_name_different_prefixes(self):
"""Test collection name creation with different prefixes"""
user = "test_user"
collection = "test_collection"
dimension = 384
doc_result = make_safe_collection_name(user, collection, dimension, "doc")
entity_result = make_safe_collection_name(user, collection, dimension, "entity")
custom_result = make_safe_collection_name(user, collection, dimension, "custom")
assert doc_result == "doc_test_user_test_collection_384"
assert entity_result == "entity_test_user_test_collection_384"
assert custom_result == "custom_test_user_test_collection_384"
doc_result = make_safe_collection_name(user, collection, "doc")
entity_result = make_safe_collection_name(user, collection, "entity")
custom_result = make_safe_collection_name(user, collection, "custom")
assert doc_result == "doc_test_user_test_collection"
assert entity_result == "entity_test_user_test_collection"
assert custom_result == "custom_test_user_test_collection"
def test_make_safe_collection_name_different_dimensions(self):
"""Test collection name creation with different dimensions"""
"""Test collection name creation - dimension handling no longer part of function"""
user = "test_user"
collection = "test_collection"
prefix = "doc"
result_128 = make_safe_collection_name(user, collection, 128, prefix)
result_384 = make_safe_collection_name(user, collection, 384, prefix)
result_768 = make_safe_collection_name(user, collection, 768, prefix)
assert result_128 == "doc_test_user_test_collection_128"
assert result_384 == "doc_test_user_test_collection_384"
assert result_768 == "doc_test_user_test_collection_768"
# With new API, dimensions are handled separately, function always returns same result
result = make_safe_collection_name(user, collection, prefix)
assert result == "doc_test_user_test_collection"
def test_make_safe_collection_name_long_names(self):
"""Test collection name creation with very long user/collection names"""
long_user = "a" * 100
long_collection = "b" * 100
result = make_safe_collection_name(
user=long_user,
collection=long_collection,
dimension=384,
prefix="doc"
)
expected = f"doc_{long_user}_{long_collection}_384"
expected = f"doc_{long_user}_{long_collection}"
assert result == expected
assert len(result) > 200 # Verify it handles long names
@ -178,20 +161,18 @@ class TestMilvusCollectionNaming:
result = make_safe_collection_name(
user="user123",
collection="collection456",
dimension=384,
prefix="doc"
)
assert result == "doc_user123_collection456_384"
assert result == "doc_user123_collection456"
def test_make_safe_collection_name_case_sensitivity(self):
"""Test that collection name creation preserves case"""
result = make_safe_collection_name(
user="TestUser",
collection="TestCollection",
dimension=384,
prefix="Doc"
)
assert result == "Doc_TestUser_TestCollection_384"
assert result == "Doc_TestUser_TestCollection"
def test_make_safe_collection_name_realistic_examples(self):
"""Test collection name creation with realistic user/collection combinations"""
@ -202,30 +183,27 @@ class TestMilvusCollectionNaming:
("user_123", "test_collection", "user_123", "test_collection"),
("αβγ-user", "测试集合", "user", "default"),
]
for user, collection, expected_user, expected_collection in test_cases:
result = make_safe_collection_name(user, collection, 384, "doc")
assert result == f"doc_{expected_user}_{expected_collection}_384"
result = make_safe_collection_name(user, collection, "doc")
assert result == f"doc_{expected_user}_{expected_collection}"
def test_make_safe_collection_name_matches_qdrant_pattern(self):
"""Test that Milvus collection names follow similar pattern to Qdrant"""
"""Test that Milvus collection names follow similar pattern to Qdrant (but without dimension in name)"""
# Qdrant uses: "d_{user}_{collection}_{dimension}" and "t_{user}_{collection}_{dimension}"
# Milvus should use: "{prefix}_{safe_user}_{safe_collection}_{dimension}"
# New Milvus API uses: "{prefix}_{safe_user}_{safe_collection}" (dimension handled separately)
user = "test.user@domain.com"
collection = "test-collection.v2"
dimension = 384
doc_result = make_safe_collection_name(user, collection, dimension, "doc")
entity_result = make_safe_collection_name(user, collection, dimension, "entity")
# Should follow the pattern but with sanitized names
assert doc_result == "doc_test_user_domain_com_test_collection_v2_384"
assert entity_result == "entity_test_user_domain_com_test_collection_v2_384"
# Verify structure matches expected pattern (may have more underscores due to sanitization)
# The important thing is that it follows prefix_user_collection_dimension structure
doc_result = make_safe_collection_name(user, collection, "doc")
entity_result = make_safe_collection_name(user, collection, "entity")
# Should follow the pattern but with sanitized names and no dimension
assert doc_result == "doc_test_user_domain_com_test_collection_v2"
assert entity_result == "entity_test_user_domain_com_test_collection_v2"
# Verify structure matches expected pattern
assert doc_result.startswith("doc_")
assert doc_result.endswith("_384")
assert entity_result.startswith("entity_")
assert entity_result.endswith("_384")
# Dimension is no longer part of the collection name

View file

@ -32,7 +32,7 @@ class TestMilvusUserCollectionIntegration:
doc_vectors.insert(vector, "test document", user, collection)
expected_collection_name = make_safe_collection_name(
user, collection, len(vector), "doc"
user, collection, "doc"
)
# Verify collection was created with correct name
@ -58,7 +58,7 @@ class TestMilvusUserCollectionIntegration:
entity_vectors.insert(vector, "test entity", user, collection)
expected_collection_name = make_safe_collection_name(
user, collection, len(vector), "entity"
user, collection, "entity"
)
# Verify collection was created with correct name
@ -89,7 +89,7 @@ class TestMilvusUserCollectionIntegration:
result = doc_vectors.search(vector, user, collection, limit=5)
# Verify search was called with correct collection name
expected_collection_name = make_safe_collection_name(user, collection, 3, "doc")
expected_collection_name = make_safe_collection_name(user, collection, "doc")
mock_client.search.assert_called_once()
search_call = mock_client.search.call_args
assert search_call[1]["collection_name"] == expected_collection_name
@ -118,7 +118,7 @@ class TestMilvusUserCollectionIntegration:
result = entity_vectors.search(vector, user, collection, limit=5)
# Verify search was called with correct collection name
expected_collection_name = make_safe_collection_name(user, collection, 3, "entity")
expected_collection_name = make_safe_collection_name(user, collection, "entity")
mock_client.search.assert_called_once()
search_call = mock_client.search.call_args
assert search_call[1]["collection_name"] == expected_collection_name
@ -142,9 +142,9 @@ class TestMilvusUserCollectionIntegration:
collection_names = set(doc_vectors.collections.values())
expected_names = {
"doc_user1_collection1_3",
"doc_user2_collection2_3",
"doc_user1_collection2_3"
"doc_user1_collection1",
"doc_user2_collection2",
"doc_user1_collection2"
}
assert collection_names == expected_names
@ -167,9 +167,9 @@ class TestMilvusUserCollectionIntegration:
collection_names = set(entity_vectors.collections.values())
expected_names = {
"entity_user1_collection1_3",
"entity_user2_collection2_3",
"entity_user1_collection2_3"
"entity_user1_collection1",
"entity_user2_collection2",
"entity_user1_collection2"
}
assert collection_names == expected_names
@ -194,10 +194,13 @@ class TestMilvusUserCollectionIntegration:
collection_names = set(doc_vectors.collections.values())
expected_names = {
"doc_test_user_test_collection_3", # 3D
"doc_test_user_test_collection_4", # 4D
"doc_test_user_test_collection_2" # 2D
"doc_test_user_test_collection", # Same name for all dimensions
"doc_test_user_test_collection", # now stored per dimension in key
"doc_test_user_test_collection" # but collection name is the same
}
# Note: Now all dimensions use the same collection name, they are differentiated by the key
assert len(collection_names) == 1 # Only one unique collection name
assert "doc_test_user_test_collection" in collection_names
assert collection_names == expected_names
@patch('trustgraph.direct.milvus_doc_embeddings.MilvusClient')
@ -220,7 +223,7 @@ class TestMilvusUserCollectionIntegration:
# Verify only one collection was created
assert len(doc_vectors.collections) == 1
expected_collection_name = "doc_test_user_test_collection_3"
expected_collection_name = "doc_test_user_test_collection"
assert doc_vectors.collections[(3, user, collection)] == expected_collection_name
@patch('trustgraph.direct.milvus_doc_embeddings.MilvusClient')
@ -233,10 +236,10 @@ class TestMilvusUserCollectionIntegration:
# Test various special character combinations
test_cases = [
("user@domain.com", "test-collection.v1", "doc_user_domain_com_test_collection_v1_3"),
("user_123", "collection_456", "doc_user_123_collection_456_3"),
("user with spaces", "collection with spaces", "doc_user_with_spaces_collection_with_spaces_3"),
("user@@@test", "collection---test", "doc_user_test_collection_test_3"),
("user@domain.com", "test-collection.v1", "doc_user_domain_com_test_collection_v1"),
("user_123", "collection_456", "doc_user_123_collection_456"),
("user with spaces", "collection with spaces", "doc_user_with_spaces_collection_with_spaces"),
("user@@@test", "collection---test", "doc_user_test_collection_test"),
]
vector = [0.1, 0.2, 0.3]
@ -250,24 +253,24 @@ class TestMilvusUserCollectionIntegration:
def test_collection_name_backward_compatibility(self):
"""Test that new collection names don't conflict with old pattern"""
# Old pattern was: {prefix}_{dimension}
# New pattern is: {prefix}_{safe_user}_{safe_collection}_{dimension}
# New pattern is: {prefix}_{safe_user}_{safe_collection}
# The new pattern should never generate names that match the old pattern
old_pattern_examples = ["doc_384", "entity_768", "doc_512"]
test_cases = [
("user", "collection", 384, "doc"),
("test", "test", 768, "entity"),
("a", "b", 512, "doc"),
("user", "collection", "doc"),
("test", "test", "entity"),
("a", "b", "doc"),
]
for user, collection, dimension, prefix in test_cases:
new_name = make_safe_collection_name(user, collection, dimension, prefix)
# New names should have at least 4 underscores (prefix_user_collection_dimension)
for user, collection, prefix in test_cases:
new_name = make_safe_collection_name(user, collection, prefix)
# New names should have at least 2 underscores (prefix_user_collection)
# Old names had only 1 underscore (prefix_dimension)
assert new_name.count('_') >= 3, f"New name {new_name} doesn't have enough underscores"
assert new_name.count('_') >= 2, f"New name {new_name} doesn't have enough underscores"
# New names should not match old pattern
assert new_name not in old_pattern_examples, f"New name {new_name} conflicts with old pattern"
@ -286,23 +289,23 @@ class TestMilvusUserCollectionIntegration:
dimension = 384
# Generate collection names
doc_name1 = make_safe_collection_name(user1, collection1, dimension, "doc")
doc_name2 = make_safe_collection_name(user2, collection2, dimension, "doc")
entity_name1 = make_safe_collection_name(user1, collection1, dimension, "entity")
entity_name2 = make_safe_collection_name(user2, collection2, dimension, "entity")
doc_name1 = make_safe_collection_name(user1, collection1, "doc")
doc_name2 = make_safe_collection_name(user2, collection2, "doc")
entity_name1 = make_safe_collection_name(user1, collection1, "entity")
entity_name2 = make_safe_collection_name(user2, collection2, "entity")
# Verify complete isolation
assert doc_name1 != doc_name2, "Document collections should be isolated"
assert entity_name1 != entity_name2, "Entity collections should be isolated"
# Verify names match expected pattern from Qdrant
# Verify names match expected pattern from new API
# Qdrant uses: d_{user}_{collection}_{dimension}, t_{user}_{collection}_{dimension}
# Milvus uses: doc_{safe_user}_{safe_collection}_{dimension}, entity_{safe_user}_{safe_collection}_{dimension}
assert doc_name1 == "doc_my_user_test_coll_1_384"
assert doc_name2 == "doc_other_user_production_data_384"
assert entity_name1 == "entity_my_user_test_coll_1_384"
assert entity_name2 == "entity_other_user_production_data_384"
# New Milvus API uses: doc_{safe_user}_{safe_collection}, entity_{safe_user}_{safe_collection}
assert doc_name1 == "doc_my_user_test_coll_1"
assert doc_name2 == "doc_other_user_production_data"
assert entity_name1 == "entity_my_user_test_coll_1"
assert entity_name2 == "entity_other_user_production_data"
# This test would have FAILED with the old implementation that used:
# - doc_384 for all document embeddings (no user/collection differentiation)