SQL completion for custom JDBC data sources knew tables but no columns: the JDBC completion model left every table's columns empty (no alias. members, no WHERE columns, no SELECT * expansion). The bridge returns a namespace's columns in one DatabaseMetaData.getColumns call (schemaColumns) and the model fills each table; a driver refusing it still completes tables.
When a proxy or tunnel dropped an idle custom JDBC connection, every later statement failed and the source still showed Connected: the cached pool was only marked closed when the IDE closed it. After a failed statement the bridge now checks Connection.isValid (new valid op); an invalid or orphaned session closes the pool so the next statement or Connect reconnects. Keep-alive for JDBC uses the same check (and its option shows in the dialog); failures log at info.
A Custom JDBC driver connects any database that ships a JDBC driver. Driver profiles (driver and dependency JARs, driver class detected from the JARs, URL template with {host}/{port}/{database}/{file} fields) are stored locally and shared by all projects; project files can't start a driver until the user approves the source. Queries run through a bundled Java 17 bridge (java/jdbc-bridge, reproducible from source) found with the existing JDK discovery. Options › Read-only calls Connection.setReadOnly(true) and the IDE rejects table edits; Cancel calls Statement.cancel() and keeps the session, with a three-second fallback that reconnects. Transactions, Explain Plan and grid edits are off for JDBC sources.
Co-authored-by: principalwater <[email protected]>