pgagroal

Appearance

Performance ​

Performance is an important goal for pgagroal and effort have been made to make pgagroal scale and use a limited number of resources.

This chapter describes pgagroal performance characteristics and provides benchmarking results compared to other PostgreSQL connection pool implementations.

Benchmarking Methodology ​

The pgbench program was used in the performance runs. All pool configurations were made with performance in mind.

The runs were performed on RHEL 7.7 / EPEL / DevTools 8 based machines on 10G network. All connection pools were the latest versions as of January 14, 2020. pgagroal was using the epoll mode of libev.

Performance Results ​

Simple Protocol ​

This run uses:

pgbench -M simple

pgbench simple

Extended Protocol ​

This run uses:

pgbench -M extended

pgbench extended

Prepared Statements ​

This run uses:

pgbench -M prepared

pgbench prepared

Read-Only Workload ​

This run uses:

pgbench -S

pgbench readonly

Performance Tuning ​

Pipeline Selection ​

Choose the appropriate pipeline for your workload:

  • Performance pipeline: Fastest option for high-throughput scenarios
  • Session pipeline: Balanced performance with full feature support
  • Transaction pipeline: Best for applications with many short transactions

See Pipelines for detailed configuration.

Connection Pool Sizing ​

Optimal pool sizing depends on your workload:

  • CPU-bound workloads: Pool size approximately equals number of CPU cores
  • I/O-bound workloads: Pool size can be higher than CPU cores
  • Mixed workloads: Start with 2x CPU cores and adjust based on monitoring

System-Level Optimizations ​

Network Configuration ​

  • Use dedicated network interfaces for database traffic
  • Configure appropriate TCP buffer sizes
  • Consider using 10G or higher network speeds for high-throughput scenarios

Memory Configuration ​

  • Enable huge pages for better memory management
  • Configure appropriate shared memory settings
  • Monitor memory usage patterns

CPU Configuration ​

  • Pin pgagroal processes to specific CPU cores if needed
  • Configure CPU governor for performance
  • Monitor CPU utilization patterns

Monitoring Performance ​

Use the following metrics to monitor pgagroal performance:

  • Connection utilization: Active vs. total connections
  • Response times: Average and percentile response times
  • Throughput: Transactions per second
  • Resource usage: CPU, memory, and network utilization

See Prometheus for detailed monitoring setup.