DBeaver UI Thread vs. JDBC Connection Pool: How Does Concurrent Query Execution Affect Latency?

Interaction of UI Thread and JDBC Pool

In DBeaver the UI thread runs the SWT event loop; when multiple SQL statements are issued, the thread serializes the submission and result‑set rendering. If the underlying JDBC connection is limited to a single physical connection per session, each statement must wait for the previous one to finish, causing queueing at the driver level. Thus UI thread contention and connection‑slot saturation compound latency.

Confirmed facts

• The UI thread cannot process more than one query submission or result‑set UI update at a time.
• DBeaver typically uses one JDBC connection per database session; there is no tunable HikariCP‑style pool in the standard editions.
• Large result sets are fully buffered in memory, which adds GC pressure and UI paint delays under concurrency.
• Enabling fetch‑on‑demand or reducing max‑rows limits reduces initial memory use but may increase round‑trips.

Configuration levers (no driver default change)

1. Set “Result Set → Fetch Size” to a modest value (e.g., 1000 rows) to avoid full buffering.
2. Enable “Edit → Preferences → Data Loading → Load data on demand” to defer fetching until UI scroll.
3. Limit “Maximum concurrent queries” per connection in the connection properties if the driver supports statement‑slot limits.
4. Close idle result sets or use separate tabs/connections for parallel work instead of reusing a single connection.

Measuring UI thread vs. pool impact

To isolate the two sources, capture a thread dump (jstack or JFR) while running N concurrent queries and compare the state of the UI thread against worker threads blocked on JDBC calls. Then measure end‑to‑end latency for a single query versus the same N queries executed concurrently, first with default fetch settings and then with fetch‑on‑demand enabled. The delta reveals whether UI rendering or connection‑slot queuing dominates.

Please provide the DBeaver edition and version (Community or Enterprise, e.g., 22.2) so we can confirm whether a shared connection pool exists and tailor the recommendation.