如何调用Cassandra自定义RequestLogger构造函数并推送Micrometer指标到Prometheus？

核心原因 #

Cassandra Java Driver初始化RequestLogger时，只会调用你已实现的带DriverContext或RequestLogFormatter参数的构造函数——Driver框架本身不知道你定义的MeterRegistry构造函数，因此不会自动触发调用。要解决这个问题，需要手动将MeterRegistry注入到自定义Logger的实例化流程中。

解决方案步骤 #

1. 重构CustomRequestLogger的构造逻辑

修改自定义Logger的构造，将MeterRegistry作为依赖整合到已被Driver调用的构造链中，同时初始化指标：

public class CustomRequestLogger extends RequestLogger {
    private final RequestLogFormatter formatter;
    private Gauge gaugeQueryExecTime;
    // 维护指标所需的字段变量
    private String currentClientId;
    private String currentQuery;
    private double currentDuration;

    // 新增接收DriverContext和MeterRegistry的构造
    public CustomRequestLogger(DriverContext context, MeterRegistry meterRegistry) {
        this(new RequestLogFormatter(context), meterRegistry);
    }

    // 整合Formatter和MeterRegistry的构造
    protected CustomRequestLogger(RequestLogFormatter formatter, MeterRegistry meterRegistry) {
        this.formatter = formatter;
        initCustomMetrics(meterRegistry);
    }

    // 初始化Gauge指标
    private void initCustomMetrics(MeterRegistry meterRegistry) {
        gaugeQueryExecTime = Gauge.builder("query.timer.duration", this, CustomRequestLogger::getCurrentDuration)
                // 添加tag关联client_id和query维度
                .tag("client_id", () -> this.currentClientId)
                .tag("query", () -> this.currentQuery)
                .register(meterRegistry);
    }

    // 重写日志回调方法，从请求日志中提取并更新字段
    @Override
    protected void onSuccess(SuccessfulRequestLog log) {
        this.currentDuration = log.getDuration().toMillis();
        this.currentClientId = log.getClientAddress().toString();
        this.currentQuery = log.getQuery();
        super.onSuccess(log);
    }

    // Gauge的取值方法
    public double getCurrentDuration() {
        return currentDuration;
    }
}

2. 在Spring配置中手动实例化CustomRequestLogger

因为你使用了Prometheus Actuator，推测是Spring Boot环境，通过配置类将MeterRegistry注入到自定义Logger中，并配置到Cassandra Driver：

如果使用原生Cassandra Driver：

@Configuration
public class CassandraDriverConfig {

    @Autowired
    private MeterRegistry meterRegistry;

    @Bean
    public CqlSession cqlSession() {
        DriverConfigLoader configLoader = DriverConfigLoader.fromClasspath("application.conf");
        DriverContext driverContext = new DefaultDriverContext(configLoader, ProgrammaticArguments.empty());

        // 手动实例化CustomRequestLogger，传入MeterRegistry
        CustomRequestLogger customLogger = new CustomRequestLogger(driverContext, meterRegistry);

        // 将自定义Logger添加到Driver的SessionFactory
        driverContext.getSessionFactoryBuilder()
                .addRequestLogger(customLogger);

        return CqlSession.builder().withConfigLoader(configLoader).build();
    }
}

如果使用Spring Data Cassandra，通过SessionBuilderCustomizer注入：

@Component
public class CassandraSessionCustomizer implements SessionBuilderCustomizer {

    @Autowired
    private MeterRegistry meterRegistry;

    @Override
    public void customize(CqlSessionBuilder sessionBuilder) {
        sessionBuilder.withSessionFactoryBuilderCustomizer(factoryBuilder -> {
            DriverContext context = factoryBuilder.getContext();
            CustomRequestLogger customLogger = new CustomRequestLogger(context, meterRegistry);
            factoryBuilder.addRequestLogger(customLogger);
        });
    }
}

额外建议 #

针对请求耗时监控，Timer指标比Gauge更适合——Timer可以自动统计请求耗时的分位数、总和、计数等维度，更符合Prometheus的监控场景。你可以将Gauge替换为Timer：

private Timer queryTimer;

private void initCustomMetrics(MeterRegistry meterRegistry) {
    queryTimer = Timer.builder("query.timer.duration")
            .register(meterRegistry);
}

@Override
protected void onSuccess(SuccessfulRequestLog log) {
    // 记录单个请求的耗时
    queryTimer.record(log.getDuration());
    // 按client_id或query维度拆分指标
    queryTimer.tagged("client_id", log.getClientAddress().toString(), "query", log.getQuery())
            .record(log.getDuration());
    super.onSuccess(log);
}

内容的提问来源于stack exchange，提问作者enthusiast