Skip to content

performance.md

Latest commit

 

History

History
64 lines (44 loc) · 2.06 KB

performance.md

File metadata and controls

64 lines (44 loc) · 2.06 KB
Raw

Performance

This document captures current performance-oriented design choices and measurement options.

Current Optimizations

  • Dispatcher interceptor fast-path (hasInterceptors) avoids loop overhead when none are registered.
  • Batch dispatch pre-sizes response list with batch size.
  • WebMVC endpoint validates payload size before JSON parse.
  • Metrics interceptor avoids per-call timer builder allocation.

Notification Throughput Strategy

Notifications can run async using jsonrpc.notification-executor-enabled=true.

Recommended production setup:

  1. Provide dedicated Executor bean for JSON-RPC notifications.
  2. Set jsonrpc.notification-executor-bean-name explicitly.
  3. Tune pool size/queue based on notification traffic profile.

Benchmarking

JMH benchmark exists in jsonrpc-core:

./gradlew :jsonrpc-core:jmh

It includes dispatcher scenarios for single success/error/invalid cases and large batch profiles (all-success, all-error, mixed, notification-only).

Quick profile (short warmup/measurement):

./gradlew :jsonrpc-core:jmhQuick

Run quick profile for a specific benchmark include pattern:

./gradlew :jsonrpc-core:jmhQuick -PjmhQuickInclude=JsonRpcDispatcherBenchmark.dispatchSingle

Practical Tuning Checklist

  • Adjust jsonrpc.max-request-bytes to realistic payload limits.
  • Adjust jsonrpc.max-batch-size to protect CPU spikes.
  • Keep include-error-data=false in production unless required.
  • Disable metrics if not needed (jsonrpc.metrics-enabled=false).
  • Use allowlist/denylist to reduce exposed method surface area.
  • Set jsonrpc.metrics-max-method-tag-values to bound method tag cardinality.
  • Enable histogram/percentiles only when needed:
    • jsonrpc.metrics-latency-histogram-enabled
    • jsonrpc.metrics-latency-percentiles

Performance Testing Guidance

For representative results, test with:

  • mixed success/error/notification traffic
  • realistic payload sizes (not only small DTOs)
  • transport-level concurrency similar to deployment
  • executor saturation and queue pressure for notifications