rusackas commented on code in PR #44552:
URL: https://github.com/apache/superset/pull/44552#discussion_r4118875136
##########
superset/charts/schemas.py:
##########
@@ -1657,6 +1657,18 @@ def rename_deprecated_fields(
)
+class ChartDataStopSchema(Schema):
Review Comment:
Added, matches `ChartDataTimingSchema`'s one-liner.
##########
superset/tasks/query_cancel.py:
##########
@@ -163,3 +172,142 @@ def cancel_chart_query(
exc_info=True,
)
return False
+
+
+def _registry_key(user_id: int, client_id: str) -> str:
+ """Cache key for a user's in-flight, cancellable chart query.
+
+ The user id is part of the key rather than a field compared after lookup,
so
+ a ``client_id`` is only ever resolvable within the namespace of the user
who
+ registered it. A caller passing somebody else's ``client_id`` gets a miss —
+ it cannot read, cancel, or overwrite another user's entry.
+ """
+ return f"chart-query-cancel:{user_id}:{client_id}"
+
+
+def _registry_ttl() -> int:
+ """How long a cancel handle stays resolvable.
+
+ A synchronous chart query cannot outlive the web request running it, so the
+ webserver timeout is the natural upper bound. Entries are discarded as soon
+ as the query returns; this TTL only bounds the leak when a worker dies
+ mid-query.
+ """
+ from flask import current_app
+
+ return int(current_app.config.get("SUPERSET_WEBSERVER_TIMEOUT", 60))
+
+
+@contextmanager
+def cancellable_chart_query(
+ client_id: "str | None", database: "Database | None"
+) -> Iterator[None]:
+ """Let the requesting user cancel this synchronous chart query by
``client_id``.
+
+ Captures the engine cancel id off the live cursor (for engines that expose
+ one before execution) and publishes it so a concurrent Stop request — which
+ lands on a different worker while this one is blocked on the query — can
kill
+ the backend session. Engines without cancel support capture nothing and the
+ query stays non-cancellable, exactly as before.
+
+ A no-op without a ``client_id``, without a database (e.g. the annotation
+ datasource, which queries Superset's own metadata DB), or for an
+ unauthenticated request — an anonymous viewer of a public dashboard has no
+ user id to scope the handle to, and an unscoped handle would be cancellable
+ by any other anonymous visitor.
+ """
+ from superset.utils.core import get_user_id
+
+ user_id = get_user_id()
+ if not client_id or database is None or user_id is None:
+ yield
+ return
+
+ # Rebound as non-optional locals: mypy does not carry the narrowing above
+ # into the nested function below.
+ owner_id: int = user_id
+ query_id: str = client_id
+ target: "Database" = database
+ database_id = target.id
+ captured = False
+
+ def _sink(cursor: Any) -> None:
+ # Republish for every cursor rather than only the first: one query
+ # object can execute several statements in turn (e.g. the grouping-sets
+ # fallback), each on whatever connection the pool hands out, and the
+ # handle must always name the session that is executing right now.
+ nonlocal captured
+ cancel_id = capture_cancel_query_id(target, cursor)
+ if cancel_id is None:
+ return
+ captured = True
+ _publish_cancel_handle(owner_id, query_id, database_id, cancel_id)
+
+ try:
+ with capture_cancel_id(_sink):
+ yield
+ finally:
+ if captured:
+ _discard_cancel_handle(owner_id, query_id)
+
+
+def _publish_cancel_handle(
+ user_id: int, client_id: str, database_id: int, cancel_query_id: str
+) -> None:
+ """Publish a cancel handle for the owning user. Best-effort."""
+ from superset.extensions import cache_manager
+
+ try:
+ cache_manager.cache.set(
+ _registry_key(user_id, client_id),
+ {"database_id": database_id, "cancel_query_id": cancel_query_id},
+ timeout=_registry_ttl(),
+ )
+ except Exception: # noqa: BLE001 pylint: disable=broad-except
+ # Forfeits cancellability for this query; never breaks its execution.
+ logger.warning("Could not publish chart query cancel handle",
exc_info=True)
+
+
+def _discard_cancel_handle(user_id: int, client_id: str) -> None:
+ """Drop a cancel handle once its query is no longer running.
Best-effort."""
+ from superset.extensions import cache_manager
+
+ try:
+ cache_manager.cache.delete(_registry_key(user_id, client_id))
+ except Exception: # noqa: BLE001 pylint: disable=broad-except
+ logger.warning("Could not discard chart query cancel handle",
exc_info=True)
+
+
+def cancel_chart_query_for_user(client_id: str) -> bool:
+ """Cancel the requesting user's running chart query, if it is cancellable.
+
+ :returns: True if the engine reported the query cancelled. False covers
both
+ "no such in-flight query for this user" and "the engine declined" — the
+ caller cannot distinguish them, which is deliberate: it keeps the
+ endpoint from confirming whether a given ``client_id`` exists.
+ """
+ from superset.daos.database import DatabaseDAO
+ from superset.extensions import cache_manager
+ from superset.utils.core import get_user_id
Review Comment:
Traced it, and these turn out to be load-bearing rather than sloppy.
test_query_cancel.py patches get_user_id/cache_manager at their defining
modules (superset.utils.core, superset.extensions), and a module-level `from
... import X` binds X once at import time, before any patch runs, so the patch
would silently never take effect. Confirmed by actually hoisting all six and
watching 4 of the 24 tests fail before reverting.
current_app was the one exception, nothing patches it, so that one's moved
to a normal top-level import. Added a comment at each remaining site explaining
why it stays inline.
##########
superset-frontend/src/explore/types.ts:
##########
@@ -61,6 +61,8 @@ export interface ChartState {
latestQueryFormData: LatestQueryFormData;
sliceFormData: QueryFormData | null;
queryController: AbortController | null;
+ /** Client-generated id of the in-flight query, used to cancel it
server-side. */
+ latestQueryId?: string;
Review Comment:
Traced this one, and I don't think it's reachable: RunQueryButton only
renders Stop while chartStatus === 'loading' (the Run/Update button otherwise),
and CHART_UPDATE_STARTED is exactly what sets latestQueryId, right as that same
loading window opens. So by the time Stop could actually be clicked,
latestQueryId always names the query that's currently loading, there's no path
where it survives to name a finished one, since finishing swaps the button away
before anything else could read it. Agreed the field would be misleading if
something else ever read it after the fact, but nothing does today, so I'd
rather not add dead-code clearing for it.
--
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]
