How to Slash API Response Times with Pfinder, JSF Async, and Set Refactoring

Introduction

During project development we often encounter slow API responses, which hurt user experience and system throughput. Improving performance requires comprehensive optimization across code, database, cache, and asynchronous processing. This article shares an optimization journey using Pfinder, JSF async calls, and code refactoring.

Current Diagnosis

UMP diagnostics show a maximum response time of 10 seconds, T99 of 1000 ms, and frequent availability drops. Pfinder diagnostics reveal three main issues:

Looping RPC 120 times = 1441 ms

Database query = 286 ms

Unknown operation > 2000 ms

Problem Location and Performance Optimization

Complete Pfinder Tracing

Integrate the Pfinder SDK:

<!-- 引用 PFinder SDK 库 -->
<dependency>
    <groupId>com.jd.pfinder</groupId>
    <artifactId>pfinder-profiler-sdk</artifactId>
    <version>1.2.2-FINAL</version>
</dependency>

Use the @PFTracing annotation to report tracing data.

A code snippet filters duplicate elements from the waveInfos list against sendDPackageCodes, building repeatResult and showPackages. Large collections (waveInfos > 3000, sendDPackageCodes > 7000) cause high latency.

Code Optimization with Set

Replacing list iteration with a Set reduces processing time from 2000 ms to 6 ms.

Resolving RPC Batch Calls with JSF Async

Compare synchronous and asynchronous approaches. Inject an asynchronous bean and configure a JSF consumer with the async="true" attribute.

// Sync bean
@Autowired
private XxxxxApi xxxxApi;

// Async bean
@Autowired
private XxxxxApi xxxxAsyncApi;

// JSF consumer configuration
<jsf:consumer id="xxx" interface="xxx" protocol="jsf" alias="xx" timeout="xxx" retries="0" check="false">
    <jsf:method name="methodName" async="true"/>
</jsf:consumer>

Use RpcContext to obtain a CompletableFuture:

public CompletableFuture<CommonDto<PageDto>> queryWaybillDetailByBusinessIdByAsync() {
    return RpcContext.getContext().asyncCall(() -> xxxxAsyncApi.method());
}

Retrieve the result with a timeout:

public <T> T getResultDefaultTimeOut(CompletableFuture<T> future) {
    try {
        return future.get(10, TimeUnit.SECONDS);
    } catch (InterruptedException | ExecutionException | TimeoutException e) {
        throw new RuntimeException(e);
    }
}

This optimization reduces RPC latency from 1400 ms to 200 ms.

Final Results and Unresolved Issues

The before‑and‑after performance improvements are illustrated below.

Conclusion

API performance optimization is a multi‑dimensional process that requires continuous diagnosis and targeted improvements across code, database, cache, and asynchronous processing. By applying the techniques described—Pfinder tracing, Set‑based collection handling, and JSF asynchronous calls—developers can significantly improve response times and enhance user experience.