danhuawang opened a new issue, #11401:
URL: https://github.com/apache/gravitino/issues/11401
### Version
main branch
### Describe what's wrong
`CALL gravitino.system.drop_catalog('gt_glue_xxx1')` fails with "Catalog
trino_connector_metalake.gt_glue_xxx1 not exists", but immediately calling
`create_catalog` for the same catalog fails with "already exists". The catalog
is stuck in an inconsistent state where it cannot be dropped or re-created.
The root cause is in `DropCatalogStoredProcedure.dropCatalog()`: it checks
catalog existence via `catalogConnectorManager.getCatalogConnector()`, which
only queries the Trino connector's **local in-memory cache** (the
`catalogConnectors` map). If the catalog exists on the Gravitino server but was
never successfully loaded into the connector (e.g., connector creation failed,
cache was evicted, or Trino was restarted), the drop incorrectly reports "not
exists" without ever consulting the Gravitino server.
The correct behavior should be: when the local connector is not found, fall
back to querying the Gravitino server to determine whether the catalog actually
exists, and if so, proceed with the server-side drop.
### Error message and/or stacktrace
```
trino> CALL gravitino.system.drop_catalog('gt_glue_xxx1');
Query 20260602_164118_00410_q78qz failed: Drop catalog failed. Catalog
trino_connector_metalake.gt_glue_xxx1 not exists.
io.trino.spi.TrinoException: Drop catalog failed. Catalog
trino_connector_metalake.gt_glue_xxx1 not exists.
at
org.apache.gravitino.trino.connector.system.storedprocedure.DropCatalogStoredProcedure.dropCatalog(DropCatalogStoredProcedure.java:123)
at
java.base/java.lang.invoke.MethodHandle.invokeWithArguments(MethodHandle.java:735)
at
java.base/java.lang.invoke.MethodHandle.invokeWithArguments(MethodHandle.java:760)
at io.trino.execution.CallTask.execute(CallTask.java:210)
...
Caused by: io.trino.spi.TrinoException: Catalog
trino_connector_metalake.gt_glue_xxx1 not exists.
at
org.apache.gravitino.trino.connector.system.storedprocedure.DropCatalogStoredProcedure.dropCatalog(DropCatalogStoredProcedure.java:99)
... 15 more
```
Immediately after:
```
trino> CALL gravitino.system.create_catalog('gt_glue_xxx1', 'glue', ...);
-- fails with: Catalog trino_connector_metalake.gt_glue_xxx1 already exists
in the server.
```
### How to reproduce
1. Deploy Gravitino + Trino with the Gravitino Trino connector (Trino 478).
2. Create a catalog `gt_glue_xxx1` via `create_catalog` stored procedure
with properties that will cause connector instantiation to fail (e.g., an
invalid placeholder like `${?aws_glue_endpoint}` for the endpoint URL). The
catalog metadata is persisted to the Gravitino server, but the Trino connector
fails to load.
3. The background scheduler (`gravitino-connector-schedule-0`) also attempts
to load the catalog and fails — so no `CatalogConnectorContext` is ever stored
in the local cache.
4. Call `CALL gravitino.system.drop_catalog('gt_glue_xxx1')` → fails with
"not exists" because `getCatalogConnector()` returns null.
5. Call `CALL gravitino.system.create_catalog('gt_glue_xxx1', ...)` → fails
with "already exists" because the Gravitino server still holds the metadata.
6. The catalog is now in a zombie state: cannot be dropped, cannot be
re-created.
### Additional context
The problematic code is in `DropCatalogStoredProcedure.java` lines 91-99:
```java
CatalogConnectorContext catalogConnector =
catalogConnectorManager.getCatalogConnector(
catalogConnectorManager.getTrinoCatalogName(metalake, catalogName));
if (catalogConnector == null) {
if (ignoreNotExist) {
return;
}
throw new TrinoException(
GravitinoErrorCode.GRAVITINO_CATALOG_NOT_EXISTS,
"Catalog " + NameIdentifier.of(metalake, catalogName) + " not
exists.");
}
```
**Suggested fix**: When `getCatalogConnector()` returns null, query the
Gravitino server directly (via the Gravitino client API) to check if the
catalog exists. If it does, issue the server-side drop. Only throw "not exists"
if the catalog is absent from both the local cache and the server.
**Workaround**: Use `CALL gravitino.system.drop_catalog('gt_glue_xxx1',
true)` (with `ignoreNotExist=true`) to skip the local cache check, though this
may not actually remove the catalog from the Gravitino server either.
--
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]
