Cross-node parallel execution
Elastic Parallel Query (EPQ) lets SQL queries run across multiple compute nodes simultaneously, delivering faster results for large-scale analytical workloads without requiring you to move or reshard data.
How it works
Without EPQ, each SQL query is processed on a single node. All data must pass through that node, creating bottlenecks in I/O, CPU, and memory as query complexity and data volume grow.
With EPQ, a query is distributed across multiple read-only nodes in parallel. Each node processes a portion of the data locally, and results are consolidated before being returned. This approach increases I/O throughput on Polar File System (PolarFS), raises CPU utilization across compute nodes, and improves memory usage across compute nodes—collectively improving query performance.
Because PolarDB for PostgreSQL (Compatible with Oracle) uses a compute-storage separation architecture, data does not need to be duplicated or resharded when nodes are added. All compute nodes share the same storage, so newly added read-only nodes are automatically included in parallel execution.
Key capabilities
Hybrid transactional and analytical processing (HTAP)
Run real-time analytical queries directly on transaction processing (TP) data without copying data to a separate system. Isolate read-only nodes for analytical workloads from those handling transactional workloads, keeping TP services unaffected.
Elastic scaling without resharding
Add read-only nodes to increase parallel query capacity based on your workload. The new nodes join the parallel execution pool automatically—no data migration or resharding required, and no data skew occurs.
Use cases
| Scenario | Description |
|---|---|
| Large-scale analytical queries | Run complex aggregations and scans on TP data in real time, without an ETL pipeline or dedicated analytical cluster |
| Workload isolation | Dedicate separate read-only nodes to analytical queries so reporting workloads do not compete with transactional traffic |
| Elastic capacity scaling | Scale out parallel query capacity by adding read-only nodes, proportional to query concurrency or data volume growth |
EPQ provides the most benefit for long-running queries that scan large amounts of data but return a small result set, such as aggregations, joins across large tables, and full-table scans. Short transactional queries that access a few rows do not benefit significantly.
What's next
For configuration details, parameter reference, and query plan examples for Elastic Parallel Query, see Elastic Parallel Query.