卓越工程之如何做好Code Review

 

- public Level queryLevel(LevelQueryRequest request) {+ public Level queryLevelWithExpireRefresh(LevelQueryRequest request) {
    Level result = levelRepository.findLevelWithoutInit(request.getId());    if (null == result || isExpired(result.getEndTime())) {        // 如果等级为空，兜底返回L0；等级已过期，实时返回默认等级，并异步刷新        if (result == null) {            result = levelRepository.buildInitLevel(request.getId(), LevelEnum.L0);        }
        //查询为空，或者已过期发送消息，刷新等级-       LevelRefreshRequest refreshRequest = buildRefreshRequest(request);-       levelWriteService.refreshLevel(message.getId(), refreshRequest);        +       RefreshMessage refreshMessage = buildRefreshMessage(request);+       refreshMessageProducer.sendMessage(refreshMessage);    }
    return result;}
- public class RefreshMessageListener extends AbstractMessageListener {+ public class RefreshMessageListener extends AbstractOrderlyMessageListener {
    @Autowired-    private LevelWriteService levelWriteService; +    private LevelWriteRegionalService levelWriteRegionalService;
    @Override    protected boolean process(String tags, String msgId, String receivedMsg) {        RefreshMessage message = JSON.parseObject(receivedMsg, RefreshMessage.class);        if (message == null || message.getId() == null) {            log.warn("message is invalid, ignored, src={}", receivedMsg);            return true;        }
        LevelRefreshRequest refreshRequest = buildRefreshRequest(message);-       levelWriteService.refreshLevel(message.getId(), refreshRequest);+       levelWriteRegionalService.refreshLevel(message.getId(), refreshRequest);
        return true;    }}

• 为什么存在等级为空的情况？

• 为什么要设计成读时写？

• 为什么不是直接计算等级，而需要用消息队列？

• 为什么要用Regional（区域化）接口和有序消息刷新等级？

CASE1：一个业务使用3个时间穿越开关

背景

private static final String CODE = "BENEFIT_TIME_THROUGH";
public Date driftedNow(String userId) {    try {        TimeMockResult<Long> result = timeThroughService.getFutureTime(CODE, userId);        if (result.isSuccess()) {            return new Date(result.getData());        }    } catch (Throwable t) {        log.error("timeThroughService error. userId={}", userId, t);    }    return new Date();}

观点1：彻底服务化

观点2：富客户端

观点3：配置统一，重复代码三处收拢为两处

总结：

CASE2：SSR（服务端渲染）API稳定性优化

背景

// 提交任务ioTaskList.stream().forEach(t -> futures.add(pool.submit(() -> t.service.invoke())));
// 阻塞获取任务结果futures.stream().forEach(f -> {    try {        result.add(f.get());    } catch (Exception e) {        log.error(e.getMessage(), e);    }});

Step1：增加固定超时控制

// 提交任务ioTaskList.stream().forEach(t -> futures.add(pool.submit(() -> t.service.invoke())));
// 阻塞获取任务结果futures.stream().forEach(f -> {    try {-       result.add(f.get());+       result.add(f.get(1000, TimeUnit.MICROSECONDS));    } catch (Exception e) {        log.error(e.getMessage(), e);    }});

Step2：自适应超时控制

public abstract class BaseService<T> implements Service {
    @Override    public T invoke(ServiceContext context) {        Entry entry = null;        try {            // 根据service类别构造降级资源            String resourceName = "RESOURCE_" + name();            entry = SphU.entry(resourceName);            try {                // 未触发降级，正常调用后端服务                return realInvoke(context);            } catch (Exception e) {                // 业务异常，记录错误日志，返回出错信息                return failureResult(context);            }        } catch (BlockException e) {            // 被降级，可以fail fast或返回兜底数据            return degradeResult(context);        } finally {            entry.exit();        }    }
    public abstract T realInvoke();}

Step3：自适应超时控制+自定义资源key

public abstract class BaseService<T> implements Service {
    @Override    public T invoke(ServiceContext context) {        Entry entry = null;        try {            // 这里的key由service实现，融合了服务类型和自定义key构造降级资源            String resourceName = "RESOURCE_" + key(context);            entry = SphU.entry(resourceName);            try {                // 未触发降级，正常调用后端服务                return realInvoke(context);            } catch (Exception e) {                // 业务异常，记录错误日志，返回出错信息                return failureResult(context);            }        } catch (BlockException e) {            // 被降级，可以fail fast或返回兜底数据            return degradeResult(context);        } finally {            entry.exit();        }    }    
    public abstract String key(ServiceContext context);
    public abstract T realInvoke();}

总结：

参考：

1. https://en.wikipedia.org/wiki/Code_review
2. https://smartbear.com/learn/code-review/agile-code-review-process/
3. Lessons From Google: How Code Reviews Build Company Culture
4. 阿里巴巴Java编码规约：https://github.com/alibaba/p3c
5. 五种Code Review反模式：https://blogs.oracle.com/javamagazine/post/five-code-review-antipatterns
6. Capers Jones对软件质量的研究分享：http://sqgne.org/presentations/2012-13/Jones-Sep-2012.pdf
7. Modern Code Review: A Case Study at Google：https://sback.it/publications/icse2018seip.pdf
8. 智能编码插件：https://developer.aliyun.com/tool/cosy