aminghadersohi commented on code in PR #40960: URL: https://github.com/apache/superset/pull/40960#discussion_r3493590034
########## superset/mcp_service/dashboard/tool/manage_native_filters.py: ########## @@ -0,0 +1,508 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. + +""" +MCP tool: manage_native_filters + +Adds, updates, removes, and reorders native filters on a dashboard by +translating high-level operations into the ``deleted`` / ``modified`` / +``reordered`` payload consumed by ``UpdateDashboardNativeFiltersCommand``. +""" + +import copy +import logging +from typing import Any, cast + +from fastmcp import Context +from superset_core.mcp.decorators import tool, ToolAnnotations + +from superset.extensions import event_logger +from superset.mcp_service.dashboard.constants import generate_id +from superset.mcp_service.dashboard.schemas import ( + FilterSelectSpec, + FilterTimeSpec, + ManageNativeFiltersRequest, + ManageNativeFiltersResponse, + NativeFilterSummary, + NativeFilterUpdateSpec, +) +from superset.mcp_service.utils import ( + escape_llm_context_delimiters, + sanitize_for_llm_context, +) +from superset.mcp_service.utils.url_utils import get_superset_base_url +from superset.utils import json + +logger = logging.getLogger(__name__) + +# Control values that map to filter_select controlValues keys. +_SELECT_CONTROL_FIELDS: dict[str, str] = { + "multi_select": "multiSelect", + "default_to_first_item": "defaultToFirstItem", + "enable_empty_filter": "enableEmptyFilter", + "sort_ascending": "sortAscending", + "search_all_options": "searchAllOptions", +} + + +class _FilterValidationError(Exception): + """Raised internally when a filter operation fails validation.""" + + +def _empty_data_mask() -> dict[str, Any]: + """Return the default data mask for a filter with no applied value.""" + return {"filterState": {"value": None}, "extraFormData": {}} + + +def _time_data_mask(default_time_range: str | None) -> dict[str, Any]: + """Build the default data mask for a time filter. + + When ``default_time_range`` is empty the filter starts unset (the empty + mask); otherwise the range is applied as both the filter state value and + the ``time_range`` extra form data. + """ + if not default_time_range: + return _empty_data_mask() + return { + "filterState": {"value": default_time_range}, + "extraFormData": {"time_range": default_time_range}, + } + + +def _validate_dataset_column(dataset_id: int, column: str) -> None: + """Validate that the dataset exists and contains the given column.""" + from superset.daos.dataset import DatasetDAO + + dataset = DatasetDAO.find_by_id(dataset_id) + if not dataset: + raise _FilterValidationError( + f"Dataset with ID {dataset_id} not found." + " Use list_datasets to get valid dataset IDs." + ) + column_names = [c.column_name for c in dataset.columns] + if column not in column_names: + raise _FilterValidationError( + f"Column '{column}' not found in dataset {dataset_id}. " + f"Available columns: {', '.join(sorted(column_names))}." + ) + + +def _build_scope( + scope_chart_ids: list[int] | None, + dashboard_chart_ids: list[int], +) -> dict[str, Any]: + """Translate scope_chart_ids into the frontend scope structure. + + The frontend expresses scope as an exclusion list, so charts NOT in + ``scope_chart_ids`` are excluded. When ``scope_chart_ids`` is None + the filter applies to all charts (empty exclusion list). + """ + if scope_chart_ids is None: + return {"rootPath": ["ROOT_ID"], "excluded": []} + unknown = sorted(set(scope_chart_ids) - set(dashboard_chart_ids)) + if unknown: + raise _FilterValidationError( + f"scope_chart_ids contains chart IDs not on the dashboard: " + f"{unknown}. Charts on this dashboard: {sorted(dashboard_chart_ids)}." + ) + excluded = sorted(set(dashboard_chart_ids) - set(scope_chart_ids)) + return {"rootPath": ["ROOT_ID"], "excluded": excluded} + + +def _build_new_filter_config( + spec: FilterSelectSpec | FilterTimeSpec, + dashboard_chart_ids: list[int], +) -> dict[str, Any]: + """Build a full native filter config dict for a new filter.""" + scope = _build_scope(spec.scope_chart_ids, dashboard_chart_ids) + filter_id = generate_id("NATIVE_FILTER") + + if isinstance(spec, FilterSelectSpec): + _validate_dataset_column(spec.dataset_id, spec.column) + control_values: dict[str, Any] = { + "multiSelect": spec.multi_select, + "defaultToFirstItem": spec.default_to_first_item, + "enableEmptyFilter": spec.enable_empty_filter, + "searchAllOptions": spec.search_all_options, + } + if spec.sort_ascending is not None: + control_values["sortAscending"] = spec.sort_ascending + return { + "id": filter_id, + "type": "NATIVE_FILTER", + "filterType": "filter_select", + "name": spec.name, + "description": spec.description, + "scope": scope, + "targets": [ + {"datasetId": spec.dataset_id, "column": {"name": spec.column}} + ], + "controlValues": control_values, + "defaultDataMask": _empty_data_mask(), + "cascadeParentIds": [], + } + + # filter_time: no dataset target, empty controlValues + return { + "id": filter_id, + "type": "NATIVE_FILTER", + "filterType": "filter_time", + "name": spec.name, + "description": spec.description, + "scope": scope, + "targets": [{}], + "controlValues": {}, + "defaultDataMask": _time_data_mask(spec.default_time_range), + "cascadeParentIds": [], + } + + +def _validate_update_type_compat( + spec: NativeFilterUpdateSpec, filter_type: str | None +) -> None: + """Reject update fields that do not apply to the filter's type.""" + select_fields_set = [ + field + for field in (*_SELECT_CONTROL_FIELDS, "dataset_id", "column") + if getattr(spec, field) is not None + ] + if filter_type != "filter_select" and select_fields_set: + raise _FilterValidationError( + f"Filter '{spec.id}' has type '{filter_type}'; fields " + f"{select_fields_set} only apply to filter_select filters." + ) + if filter_type != "filter_time" and spec.default_time_range is not None: + raise _FilterValidationError( + f"Filter '{spec.id}' has type '{filter_type}'; default_time_range " + "only applies to filter_time filters." + ) + + +def _merge_target(spec: NativeFilterUpdateSpec, merged: dict[str, Any]) -> None: + """Merge dataset_id / column changes into the filter's first target.""" + targets = merged.get("targets") or [{}] + target = dict(targets[0]) if targets else {} + dataset_id = ( + spec.dataset_id if spec.dataset_id is not None else target.get("datasetId") + ) + column = ( + spec.column + if spec.column is not None + else (target.get("column") or {}).get("name") + ) + if dataset_id is None or not column: + raise _FilterValidationError( + f"Filter '{spec.id}' is missing a dataset or column target; " + "provide both dataset_id and column to set the target." + ) + _validate_dataset_column(dataset_id, column) + target["datasetId"] = dataset_id + target["column"] = {"name": column} + merged["targets"] = [target] + + +def _merge_filter_update( + spec: NativeFilterUpdateSpec, + existing: dict[str, Any], + dashboard_chart_ids: list[int], +) -> dict[str, Any]: + """Merge a partial update into an existing filter config. + + Returns a FULL filter config (the backend command substitutes whole + entries, it does not merge deltas). + """ + merged = copy.deepcopy(existing) + _validate_update_type_compat(spec, merged.get("filterType")) + + if spec.name is not None: + merged["name"] = spec.name + if spec.description is not None: + merged["description"] = spec.description + if spec.scope_chart_ids is not None: + merged["scope"] = _build_scope(spec.scope_chart_ids, dashboard_chart_ids) + if spec.dataset_id is not None or spec.column is not None: + _merge_target(spec, merged) + + control_values = dict(merged.get("controlValues") or {}) + for field, control_key in _SELECT_CONTROL_FIELDS.items(): + value = getattr(spec, field) + if value is not None: + control_values[control_key] = value + merged["controlValues"] = control_values + + if spec.default_time_range is not None: + merged["defaultDataMask"] = _time_data_mask(spec.default_time_range) + + return merged + + +def _filter_summary(conf: dict[str, Any]) -> NativeFilterSummary: + """Summarize a filter config for the response. + + Returns the id, name, filterType, and non-empty targets; empty target + entries (e.g. for time filters) are dropped so the summary only lists + real dataset/column targets. The user-controlled ``name`` and ``targets`` + come from dashboard metadata and are wrapped as untrusted content before + being exposed to LLM context (mirroring the get_dashboard_info read path). + The operational ``id`` and ``filter_type`` fields are delimiter-escaped + (not wrapped) so the LLM can pass them back verbatim in subsequent calls + while any embedded delimiter tokens are neutralized. + """ + name = conf.get("name") + targets = [t for t in (conf.get("targets") or []) if t] + return NativeFilterSummary( + id=escape_llm_context_delimiters(conf.get("id")), + name=sanitize_for_llm_context(name, field_path=("name",)) + if name is not None + else None, + filter_type=escape_llm_context_delimiters(conf.get("filterType")), + targets=cast( + list[dict[str, Any]], + sanitize_for_llm_context( + targets, + field_path=("targets",), + excluded_field_names=frozenset(), + ), + ), + ) + + +def _current_native_filter_config(dashboard: Any) -> list[dict[str, Any]]: + """Return the dashboard's existing native filter configuration. + + ``json_metadata`` may be missing, invalid JSON, or parse to a non-dict + (e.g. a legacy ``"[]"`` payload); all of those degrade to an empty list + rather than raising. + """ + try: + metadata = json.loads(dashboard.json_metadata or "{}") + except (json.JSONDecodeError, TypeError): + metadata = {} + if not isinstance(metadata, dict): + return [] + config = metadata.get("native_filter_configuration") + if not isinstance(config, list): + return [] + # Drop malformed (non-dict) entries so downstream conf["id"] / conf.get(...) + # cannot raise on corrupt metadata. + return [item for item in config if isinstance(item, dict)] + + +def _build_native_filters_payload( # noqa: C901 + request: ManageNativeFiltersRequest, + current_config: list[dict[str, Any]], + dashboard_chart_ids: list[int], +) -> tuple[dict[str, Any], list[str], list[str]]: + """Translate tool operations into the command payload. + + Returns ``(payload, added_filter_ids, updated_filter_ids)`` where the + payload has the ``deleted`` / ``modified`` / ``reordered`` shape expected + by ``UpdateDashboardNativeFiltersCommand``. + """ + current_by_id = {conf["id"]: conf for conf in current_config if conf.get("id")} + + unknown_removals = [fid for fid in request.remove if fid not in current_by_id] + if unknown_removals: + raise _FilterValidationError( + f"Cannot remove filters that do not exist on the dashboard: " + f"{unknown_removals}. Existing filter IDs: " + f"{sorted(current_by_id)}." + ) + + removed_ids = set(request.remove) + modified: list[dict[str, Any]] = [] + updated_filter_ids: list[str] = [] + + update_ids = [update_spec.id for update_spec in request.update] + duplicate_updates = sorted({fid for fid in update_ids if update_ids.count(fid) > 1}) + if duplicate_updates: + raise _FilterValidationError( + f"update contains duplicate filter IDs: {duplicate_updates}. " + "Provide at most one update per filter." + ) + + for update_spec in request.update: + if update_spec.id in removed_ids: + raise _FilterValidationError( + f"Filter '{update_spec.id}' cannot be both updated and removed." + ) + existing = current_by_id.get(update_spec.id) + if existing is None: + raise _FilterValidationError( + f"Cannot update filter '{update_spec.id}': not found on the " + f"dashboard. Existing filter IDs: {sorted(current_by_id)}." + ) + modified.append( + _merge_filter_update(update_spec, existing, dashboard_chart_ids) + ) + updated_filter_ids.append(update_spec.id) + + added_filter_ids: list[str] = [] + for new_spec in request.add: + config = _build_new_filter_config(new_spec, dashboard_chart_ids) + modified.append(config) + added_filter_ids.append(config["id"]) + + payload: dict[str, Any] = {} + if request.remove: + payload["deleted"] = list(request.remove) + if modified: + payload["modified"] = modified + + if request.reorder is not None: + # The DAO drops any surviving filter that is absent from the + # reordered list, so require a complete ordering of surviving + # pre-existing filters. Newly added filters are appended + # automatically by the DAO and may be omitted. + surviving_ids = set(current_by_id) - removed_ids + reorder_ids = [fid for fid in request.reorder if fid not in added_filter_ids] + if len(set(request.reorder)) != len(request.reorder): + raise _FilterValidationError("reorder contains duplicate filter IDs.") + missing = sorted(surviving_ids - set(reorder_ids)) + unknown = sorted(set(reorder_ids) - surviving_ids) + if missing or unknown: + raise _FilterValidationError( + "reorder must list every remaining filter exactly once. " + f"Missing: {missing}. Unknown: {unknown}. " + f"Remaining filter IDs: {sorted(surviving_ids)}." + ) + payload["reordered"] = list(request.reorder) + + return payload, added_filter_ids, updated_filter_ids + + +@tool( + tags=["mutate"], + class_permission_name="Dashboard", + method_permission_name="write", + annotations=ToolAnnotations( + title="Manage dashboard native filters", + readOnlyHint=False, + destructiveHint=True, + ), +) +def manage_native_filters( + request: ManageNativeFiltersRequest, ctx: Context +) -> ManageNativeFiltersResponse: + """ + Add, update, remove, and reorder native filters on a dashboard. + + Supported filter types for new filters: filter_select (dropdown backed + by a dataset column) and filter_time (time range). Other filter types + (numerical range, time column, time grain) are not yet supported. + Filter IDs are generated by the server and returned in the response. + """ + from superset.commands.dashboard.exceptions import ( + DashboardForbiddenError, + DashboardInvalidError, + DashboardNativeFiltersUpdateFailedError, + DashboardNotFoundError, + ) + from superset.commands.dashboard.update import ( + UpdateDashboardNativeFiltersCommand, + ) + from superset.commands.exceptions import TagForbiddenError + from superset.daos.dashboard import DashboardDAO + + try: + with event_logger.log_context(action="mcp.manage_native_filters.validation"): + dashboard = DashboardDAO.find_by_id(request.dashboard_id) + if not dashboard: + return ManageNativeFiltersResponse( + error=( + f"Dashboard with ID {request.dashboard_id} not found." + " Use list_dashboards to get valid dashboard IDs." + ), + ) + + current_config = _current_native_filter_config(dashboard) + dashboard_chart_ids = [slc.id for slc in dashboard.slices] + + try: + payload, added_ids, updated_ids = _build_native_filters_payload( + request, current_config, dashboard_chart_ids + ) Review Comment: The RBAC gate (`class_permission_name="Dashboard", method_permission_name="write"`) runs before the tool body, ensuring the caller has Dashboard write permission. Per-object ownership is then enforced by `UpdateDashboardNativeFiltersCommand.validate()`. The window between these two is reading `dashboard.json_metadata` and `dashboard.slices`, which are the same fields exposed by `get_dashboard_info` — a read tool available to anyone with Dashboard read access. If the concern is that a Dashboard class-write user (without ownership of the specific dashboard) can see filter IDs via error messages, that's a narrower privilege than read access to the dashboard itself. Escalating to @aminghadersohi to confirm whether this is within the intended threat model. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
