Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

修复记忆提取的 URI 异常隔离并统一只新增条目处理 (#6759)

DuTao3196b2129773
diff --git a/openviking/session/memory/dataclass.py b/openviking/session/memory/dataclass.py
index 269cb554b..12c7685ee 100644
--- a/openviking/session/memory/dataclass.py
+++ b/openviking/session/memory/dataclass.py
@@ -182,6 +182,7 @@ class MemoryOperationSkipCode(str, Enum):
AMBIGUOUS_TARGET = "ambiguous_target"
NO_WRITABLE_TARGET = "no_writable_target"
PAGE_ID_TYPE_MISMATCH = "page_id_type_mismatch"
+ INVALID_URI_TEMPLATE_VARS = "invalid_uri_template_vars"
class MemoryOperationSkip(BaseModel):
diff --git a/openviking/session/memory/extract_loop.py b/openviking/session/memory/extract_loop.py
index 9ebe459cc..0e61b8cd5 100644
--- a/openviking/session/memory/extract_loop.py
+++ b/openviking/session/memory/extract_loop.py
@@ -409,16 +409,12 @@ class ExtractLoop:
# retry would only hit the budget again, and replaying the
# truncated output just grows the prompt. Fail fast with the
# dedicated error recorded by _call_llm().
- raise RuntimeError(
- f"Memory extraction aborted: {parse_error or failure_kind}"
- )
+ raise RuntimeError(f"Memory extraction aborted: {parse_error or failure_kind}")
if self._format_retry_count == 0:
self._format_retry_count += 1
max_iterations += 1
- retry_reason = (
- "refusal_text" if failure_kind == "refusal_text" else "format_retry"
- )
+ retry_reason = "refusal_text" if failure_kind == "refusal_text" else "format_retry"
tracer.info(f"Extended max_iterations to {max_iterations} for {retry_reason}")
self._add_format_error_message(messages)
@@ -500,9 +496,7 @@ class ExtractLoop:
"""
usage = getattr(response, "usage", None)
completion_tokens = usage.get("completion_tokens") if isinstance(usage, dict) else None
- details = (
- f"finish_reason='length', output cap={self._effective_max_output_tokens} tokens"
- )
+ details = f"finish_reason='length', output cap={self._effective_max_output_tokens} tokens"
if completion_tokens:
details += f", completion_tokens={completion_tokens}"
return (
@@ -555,13 +549,6 @@ class ExtractLoop:
def _build_resolution_repair_instruction(self, issues: List[Dict[str, Any]]) -> str:
return self._output_protocol.render_resolution_repair(issues)
- @staticmethod
- def _is_event_schema(schema: Any) -> bool:
- return (
- getattr(schema, "memory_type", None) == "events"
- and getattr(schema, "operation_mode", None) == "add_only"
- )
-
def _uri_belongs_to_schema(self, uri: str, schema: Any) -> Optional[bool]:
"""Return a path-based ownership result without trusting file metadata."""
render_directories = getattr(self._isolation_handler, "render_schema_directories", None)
@@ -770,43 +757,40 @@ class ExtractLoop:
if value is None:
continue
items = value if isinstance(value, list) else [value]
- is_event = self._is_event_schema(schema)
+ is_add_only = getattr(schema, "operation_mode", "upsert") == "add_only"
for item_index, item in enumerate(items):
requested_page_id = dict(item).get("page_id")
- entries.append((schema_index, item_index, requested_page_id, is_event))
+ entries.append((schema_index, item_index, requested_page_id, is_add_only))
assigned_page_ids: Dict[Tuple[int, int], int] = {}
page_id_assignments: Dict[int, List[int]] = {}
page_id_allocator = None
- # Preserve non-Event behavior where possible. Events are always new and therefore
- # allocate after every other memory type has claimed its response-local ID.
- for event_phase in (False, True):
- for schema_index, item_index, requested_page_id, is_event in entries:
- if is_event != event_phase:
- continue
- existing_uri = None
- if requested_page_id is not None and page_id_map is not None:
- existing_uri = page_id_map.resolve(requested_page_id)
- if not is_event and existing_uri is not None:
- page_id = requested_page_id
- else:
- if page_id_allocator is None:
- allocator_factory = getattr(
- page_id_map,
- "new_page_id_allocator",
- None,
- )
- page_id_allocator = (
- allocator_factory()
- if callable(allocator_factory)
- else ResponsePageIdAllocator()
- )
- page_id = page_id_allocator.allocate(requested_page_id)
+ # Keep upsert IDs stable before allocating IDs for add_only operations.
+ entries.sort(key=lambda entry: entry[3])
+ for schema_index, item_index, requested_page_id, is_add_only in entries:
+ existing_uri = None
+ if not is_add_only and requested_page_id is not None and page_id_map is not None:
+ existing_uri = page_id_map.resolve(requested_page_id)
+ if existing_uri is not None:
+ page_id = requested_page_id
+ else:
+ if page_id_allocator is None:
+ allocator_factory = getattr(
+ page_id_map,
+ "new_page_id_allocator",
+ None,
+ )
+ page_id_allocator = (
+ allocator_factory()
+ if callable(allocator_factory)
+ else ResponsePageIdAllocator()
+ )
+ page_id = page_id_allocator.allocate(requested_page_id)
- assigned_page_ids[(schema_index, item_index)] = page_id
- if requested_page_id is not None:
- page_id_assignments.setdefault(requested_page_id, []).append(page_id)
+ assigned_page_ids[(schema_index, item_index)] = page_id
+ if requested_page_id is not None:
+ page_id_assignments.setdefault(requested_page_id, []).append(page_id)
return assigned_page_ids, page_id_assignments
@@ -828,6 +812,7 @@ class ExtractLoop:
for schema_index, schema in enumerate(schemas):
memory_type = schema.memory_type
+ is_add_only = getattr(schema, "operation_mode", "upsert") == "add_only"
value = getattr(operations, memory_type, None)
if value is None:
continue
@@ -870,7 +855,6 @@ class ExtractLoop:
requested_page_id = item_dict.pop("page_id", None)
page_id = assigned_page_ids[(schema_index, item_index)]
- is_event = self._is_event_schema(schema)
resolved_op = ResolvedOperation(
old_memory_file_content=None,
memory_fields=item_dict,
@@ -880,16 +864,14 @@ class ExtractLoop:
resolution_skip=identity_resolution_skip,
)
- if is_event:
- # Event page_ids are temporary link anchors. They must never select an
- # existing memory file; URI collision handling remains unchanged downstream.
- resolved_op.uris = self._isolation_handler.calculate_memory_uris(
- memory_type_schema=schema,
- operation=resolved_op,
- extract_context=self._extract_context,
- )
- elif requested_page_id is not None and page_id_map is not None:
- resolved_uri = page_id_map.resolve(requested_page_id)
+ try:
+ resolved_uri = None
+ if (
+ not is_add_only
+ and requested_page_id is not None
+ and page_id_map is not None
+ ):
+ resolved_uri = page_id_map.resolve(requested_page_id)
if resolved_uri:
old_content = self.context_provider.read_file_contents.get(resolved_uri)
belongs_to_schema = self._uri_belongs_to_schema(resolved_uri, schema)
@@ -940,11 +922,18 @@ class ExtractLoop:
operation=resolved_op,
extract_context=self._extract_context,
)
- else:
- resolved_op.uris = self._isolation_handler.calculate_memory_uris(
- memory_type_schema=schema,
- operation=resolved_op,
- extract_context=self._extract_context,
+ except ValueError as exc:
+ resolved_op.uris = []
+ resolved_op.resolution_skip = MemoryOperationSkip(
+ reason_code=MemoryOperationSkipCode.INVALID_URI_TEMPLATE_VARS,
+ reason=str(exc),
+ )
+ logger.warning(
+ "Skipping memory operation with unrenderable URI: "
+ "memory_type=%s page_id=%s error=%s",
+ memory_type,
+ page_id,
+ exc,
)
upsert_operations.append(resolved_op)
diff --git a/openviking/session/memory/schema_model_generator.py b/openviking/session/memory/schema_model_generator.py
index 52dcb5c46..82bb213b3 100644
--- a/openviking/session/memory/schema_model_generator.py
+++ b/openviking/session/memory/schema_model_generator.py
@@ -147,9 +147,11 @@ class SchemaModelGenerator:
page_id_json_schema = {"type": "integer"}
page_id_description = "Temporary page_id for identifying the target memory item."
- if memory_type.memory_type == "events" and memory_type.operation_mode == "add_only":
+ if memory_type.operation_mode == "add_only":
page_id_json_schema["minimum"] = 100
- page_id_description = "Unique page_id for this new event; it MUST be at least 100."
+ page_id_description = (
+ "Unique page_id for this new memory item; it MUST be at least 100."
+ )
field_definitions["page_id"] = (
Annotated[int, WithJsonSchema(page_id_json_schema)],
@@ -202,7 +204,7 @@ class SchemaModelGenerator:
if not isinstance(data, dict):
return data
try:
- is_new = int(data.get("page_id")) >= 100
+ is_new = memory_type.operation_mode == "add_only" or int(data.get("page_id")) >= 100
except (TypeError, ValueError):
return data
missing = [name for name in required_on_create if data.get(name) is None]
diff --git a/tests/session/memory/test_memory_react.py b/tests/session/memory/test_memory_react.py
index 6b79d3331..16b862d66 100644
--- a/tests/session/memory/test_memory_react.py
+++ b/tests/session/memory/test_memory_react.py
@@ -13,6 +13,7 @@ from pydantic import BaseModel, Field
from openviking.session.memory.dataclass import (
MemoryField,
MemoryFile,
+ MemoryOperationSkipCode,
MemoryTypeSchema,
ResolvedOperation,
ResolvedOperations,
@@ -196,6 +197,44 @@ class TestExtractLoopFinalJsonRetry:
extract_loop._isolation_handler = MemoryIsolationHandler(ctx, extract_context)
return extract_loop
+ @pytest.mark.asyncio
+ async def test_unrenderable_uri_skips_only_bad_item_and_reports_it(self):
+ class EventItem(BaseModel):
+ event_name: str | None = None
+
+ class Operations(BaseModel):
+ events: list[EventItem]
+
+ schema = MemoryTypeSchema(
+ memory_type="events",
+ operation_mode="add_only",
+ directory="viking://user/{{ user_space }}/memories/events",
+ filename_template="{{ event_name }}.md",
+ fields=[],
+ )
+ loop = self._invalid_peer_resolution_loop()
+ loop.context_provider.get_memory_schemas.return_value = [schema]
+ loop._extract_context.page_id_map = PageIdMap()
+ resolved, _ = await loop.resolve_operations(
+ Operations(events=[EventItem(), EventItem(event_name="kickoff")])
+ )
+
+ updater = MemoryUpdater(registry=MagicMock())
+ updater._get_viking_fs = MagicMock(return_value=MagicMock())
+ updater._apply_upsert = AsyncMock()
+ updater._sync_resource_refs_for_result = AsyncMock()
+ updater._vectorize_memories = AsyncMock()
+ updater.generate_overview = AsyncMock()
+ result = await updater.apply_operations(operations=resolved, ctx=loop.ctx)
+
+ assert result.written_uris == ["viking://user/user_a/memories/events/kickoff.md"]
+ assert result.errors == []
+ assert len(result.skipped_operations) == 1
+ assert result.skipped_operations[0].reason_code == (
+ MemoryOperationSkipCode.INVALID_URI_TEMPLATE_VARS
+ )
+ assert "event_name" in result.skipped_operations[0].reason
+
@pytest.mark.asyncio
async def test_existing_page_id_keeps_write_target_without_invalid_peer_metadata(self):
class PreferenceItem(BaseModel):
diff --git a/tests/session/memory/test_schema_models.py b/tests/session/memory/test_schema_models.py
index 50aa93eab..09cbafdf2 100644
--- a/tests/session/memory/test_schema_models.py
+++ b/tests/session/memory/test_schema_models.py
@@ -85,7 +85,6 @@ class TestSchemaModelGenerator:
"""Create a registry with real schemas."""
return get_default_registry()
-
def test_peer_enabled_false_omits_peer_id_field(self):
memory_type = MemoryTypeSchema(
memory_type="cases",
@@ -197,7 +196,6 @@ class TestSchemaModelGenerator:
assert "field1" in model.model_fields
assert "field2" in model.model_fields
-
def test_page_id_field_is_emitted_before_mutable_content(self, registry_with_sample):
"""page_id should appear before mutable fields so the model anchors target page first."""
from unittest.mock import patch
@@ -295,8 +293,10 @@ class TestSchemaModelGenerator:
assert "MUST be at least 100" in schema["properties"]["page_id"]["description"]
# Runtime resolution normalizes bad model output instead of dropping the event.
assert model.model_validate({"page_id": 5, "event_name": "demo"}).page_id == 5
+ with pytest.raises(ValueError, match="new memory item requires fields: event_name"):
+ model.model_validate({"page_id": 5, "event_name": None})
- def test_non_event_add_only_page_id_schema_is_unchanged(self):
+ def test_add_only_page_id_schema_requires_new_page_range(self):
memory_type = MemoryTypeSchema(
memory_type="trajectories",
operation_mode="add_only",
@@ -308,14 +308,10 @@ class TestSchemaModelGenerator:
model = SchemaModelGenerator([memory_type]).create_flat_data_model(memory_type)
page_id_schema = model.model_json_schema()["properties"]["page_id"]
- assert "minimum" not in page_id_schema
- assert page_id_schema["description"] == (
- "Temporary page_id for identifying the target memory item."
- )
+ assert page_id_schema["minimum"] == 100
+ assert "MUST be at least 100" in page_id_schema["description"]
- def test_existing_patch_can_omit_immutable_fields_but_new_item_cannot(
- self, real_registry
- ):
+ def test_existing_patch_can_omit_immutable_fields_but_new_item_cannot(self, real_registry):
entities = real_registry.get("entities")
model = SchemaModelGenerator([entities]).create_flat_data_model(entities)
@@ -359,14 +355,10 @@ class TestSchemaModelGenerator:
"entities": [
{
"page_id": 2,
- "content": {
- "blocks": [{"search": "一周学习计划", "replace": ""}]
- },
+ "content": {"blocks": [{"search": "一周学习计划", "replace": ""}]},
}
],
- "delete_ids": [
- {"delete_page_id": 1, "replacement_page_id": None}
- ],
+ "delete_ids": [{"delete_page_id": 1, "replacement_page_id": None}],
}
)
@@ -511,5 +503,6 @@ class TestWikiLink:
assert link_type_schema["type"] == "string"
assert "enum" not in link_type_schema
+
class TestIntegration:
"""Integration tests for the complete schema system."""
diff --git a/tests/unit/session/memory/test_extract_loop_match_text.py b/tests/unit/session/memory/test_extract_loop_match_text.py
index 0b6446751..bb473e4ee 100644
--- a/tests/unit/session/memory/test_extract_loop_match_text.py
+++ b/tests/unit/session/memory/test_extract_loop_match_text.py
@@ -281,7 +281,7 @@ class TestResolveOperations:
assert operation.resolution_skip is None
@pytest.mark.asyncio
- async def test_non_event_add_only_resolution_is_unchanged(self):
+ async def test_add_only_resolution_never_reuses_existing_page(self):
schema = MemoryTypeSchema(
memory_type="trajectories",
directory="viking://user/{{ user_space }}/memories/trajectories",
@@ -290,9 +290,14 @@ class TestResolveOperations:
fields=[],
)
trajectory_uri = "viking://user/alice/memories/trajectories/demo.md"
+ old_uri = "viking://user/alice/memories/trajectories/old.md"
+ page_id_map = PageIdMap()
+ old_page_id = page_id_map.get_page_id(old_uri)
context_provider = Mock()
context_provider.get_memory_schemas.return_value = [schema]
- context_provider.read_file_contents = {}
+ context_provider.read_file_contents = {
+ old_uri: MemoryFile(uri=old_uri, memory_type="trajectories", content="old trajectory")
+ }
isolation_handler = Mock()
isolation_handler.get_read_scope.return_value = None
isolation_handler.fill_identity_fields.side_effect = lambda item, **kwargs: item
@@ -304,15 +309,17 @@ class TestResolveOperations:
context_provider=context_provider,
isolation_handler=isolation_handler,
)
- loop._extract_context = SimpleNamespace(page_id_map=PageIdMap())
+ loop._extract_context = SimpleNamespace(page_id_map=page_id_map)
operations, _ = await loop.resolve_operations(
- AttrDict(trajectories=[{"trajectory_name": "demo", "page_id": 105}])
+ AttrDict(trajectories=[{"trajectory_name": "demo", "page_id": old_page_id}])
)
operation = operations.upsert_operations[0]
- assert operation.page_id == 105
+ assert operation.page_id == 100
assert operation.uris == [trajectory_uri]
+ assert operation.old_memory_file_content is None
+ assert page_id_map.resolve(old_page_id) == old_uri
@pytest.mark.asyncio
async def test_new_page_ids_are_unique_across_memory_types(self):
@@ -574,11 +581,7 @@ class TestResolveOperations:
)
operations, _ = await loop.resolve_operations(
- AttrDict(
- entities=[
- {"name": "Ignored", "owner": "Bob", "count": 3, "page_id": 7}
- ]
- )
+ AttrDict(entities=[{"name": "Ignored", "owner": "Bob", "count": 3, "page_id": 7}])
)
operation = operations.upsert_operations[0]
@@ -1127,9 +1130,7 @@ class TestResolutionRepair:
context_provider.get_memory_schemas.return_value = [event_schema, profile_schema]
context_provider.get_output_language.return_value = "zh-CN"
context_provider.get_tools.return_value = []
- context_provider.get_extract_context.return_value = SimpleNamespace(
- page_id_map=PageIdMap()
- )
+ context_provider.get_extract_context.return_value = SimpleNamespace(page_id_map=PageIdMap())
context_provider.prefetch = AsyncMock(return_value=[])
context_provider.read_file_contents = {}
context_provider.instruction.return_value = "base instruction"
@@ -1351,7 +1352,7 @@ class TestFinalOperationsHydration:
)
context_provider = Mock()
- schema = SimpleNamespace(memory_type="experiences", fields=[])
+ schema = MemoryTypeSchema(memory_type="experiences", fields=[])
context_provider.get_memory_schemas.return_value = [schema]
context_provider.get_output_language.return_value = "zh-CN"
context_provider.get_tools.return_value = []