Repository navigation
Description
Tracking issue for the CQL-specific work in Data Studio / MCP, split into what already landed in #214 and what remains.
Landed in #214 (a4863e4)
sqlparser has no Cassandra dialect, so two independent gates rejected every CQL-specific construct (PER PARTITION LIMIT, ALLOW FILTERING, USING TTL, token(...) ranges) with sql parser error:
| Gate | Call site | Fix |
|---|---|---|
| MCP/AI risk classification | capabilities/sql_write.rs::classify_sql |
CQL short-circuits to a keyword classifier (SELECT→Read, INSERT/UPDATE→Write, DELETE/TRUNCATE→Delete, CREATE/ALTER/DROP→Ddl, else Other) plus a literal- and comment-aware multi-statement scan. Ambiguous scripts classify as Other, so a destructive statement can never ride a read-only grant. |
| Data Studio grid wrapper | database/sql_service.rs::wrap_query |
CQL passes through unparsed. A grid sort/filter/paging request returns an explicit CQL-semantics error instead of a parser failure. |
Root cause that made both fixes unreachable: adapter_db_type() returned "generic" for every JdbcBridge connection. It now reads the adapter's real type via the new DatabaseType::slug() — this also fixed dialect selection for all other JDBC engines (ClickHouse, Hive, Databricks, …) that were being classified under GenericDialect.
Covered by tests: keyword buckets, leading comments, semicolons inside string literals ('a;b', '' escapes), trailing comments, and destructive multi-statement tails.
Remaining work
1. Grid affordances for CQL tables
The Data Studio table grid exposes column sort / filter / pagination, none of which translate faithfully to CQL:
- ORDER BY is legal only on clustering keys, in the table's declared clustering order
- there is no OFFSET
- filtering non-key columns requires
ALLOW FILTERING(a full scan) or a secondary index
Today these requests return a clear error. Options worth considering: hide the sort/filter controls for CQL connections, or derive a valid ORDER BY from system_schema.columns.kind when the requested column is a clustering key.
2. CQL statement awareness in editor tooling
Completion, the AI permission chain and any sqlparser-backed highlighting still treat CQL as generic SQL. Splitting statements on ; inside the editor is likewise unaware of CQL literals.
3. Secondary-index visibility
list_indexes returns [] for CQL because the wrapper's DatabaseMetaData does not expose system_schema.indexes. A direct system_schema query (the same approach used for tables/columns) would surface user-created indexes such as users_city_idx.
4. Upstream wrapper gaps worth reporting to ing-bank
getViewsthrowsMethod was called on a closed ResultSeton empty results (worked around in feat(db): add Apache Cassandra and ScyllaDB support #214 by returning[]for CQL)getImportedKeys/getExportedKeysthrowdoes not support this method(worked around the same way)
Verification used
Live Cassandra 4.1.12 and ScyllaDB via the event-search bastion stack, plus a tunnel_test keyspace with 46 users, 1157 hourly metric rows, 137 orders and 24 static-column rows. Query matrix exercised through both the desktop app and the MCP bridge.
Activity
Blankll commentedon Nov 18, 2026
Migrating to geek-fun/data-studio-agent#39 — the CQL gaps that live in the framework (SqlDatabase whitelist → fixed by geek-fun/data-studio-agent#38, grid affordances, editor awareness, secondary-index visibility) belong to the agent-framework repo where the capability model is defined. sqlkit-side fixes from this tracking (system_schema metadata, CQL risk classification, FK/views empty lists) are in #214.