Reporting queries timeout

The reporting service intermittently throws TimeoutException: could not acquire a connection from the pool within 30000ms. It correlates with mornings when analysts run dashboards. The HikariCP pool size is 20. Dashboards: pool.active sits at 20 during incidents while pool.pending (waiters) spikes; the database itself reports only ~30% of its max_connections used and is healthy. A change last week added an export endpoint that streams a large result set to the client while holding the DB connection open for the whole download. Triage and fix.