当前位置：首页 > web >正文

Prometheus + Grafana + Micrometer 监控方案详解

web 2025/7/30 16:30:56

这套组合是当前Java生态中最流行的监控解决方案之一，特别适合云原生环境下的微服务应用监控。下面我将从技术实现到最佳实践进行全面解析。

一、技术栈组成与协作

1. 组件分工

组件	角色	关键能力
Micrometer	应用指标门面(Facade)	统一指标采集API，对接多种监控系统
Prometheus	时序数据库+采集器	指标存储、查询、告警规则处理
Grafana	可视化平台	仪表盘展示、数据可视化分析

2. 数据流动

二、Micrometer 集成实践

1. Spring Boot 配置

Maven依赖：

<dependency><groupId>io.micrometer</groupId><artifactId>micrometer-registry-prometheus</artifactId>
</dependency>
<dependency><groupId>org.springframework.boot</groupId><artifactId>spring-boot-starter-actuator</artifactId>
</dependency>

application.yml配置：

management:endpoints:web:exposure:include: health,info,prometheusmetrics:export:prometheus:enabled: truetags:application: ${spring.application.name} # 统一添加应用标签

2. 自定义指标示例

业务指标采集：

@Service
public class OrderService {private final Counter orderCounter;private final Timer orderProcessingTimer;public OrderService(MeterRegistry registry) {// 创建计数器orderCounter = Counter.builder("orders.total").description("Total number of orders").tag("type", "online").register(registry);// 创建计时器orderProcessingTimer = Timer.builder("orders.processing.time").description("Order processing time").publishPercentiles(0.5, 0.95) // 50%和95%分位.register(registry);}public void processOrder(Order order) {// 方法1: 手动计时long start = System.currentTimeMillis();try {// 业务逻辑...orderCounter.increment();} finally {long duration = System.currentTimeMillis() - start;orderProcessingTimer.record(duration, TimeUnit.MILLISECONDS);}// 方法2: 使用Lambda自动计时orderProcessingTimer.record(() -> {// 业务逻辑...orderCounter.increment();});}
}

三、Prometheus 配置优化

1. 抓取配置示例

# prometheus.yml
global:scrape_interval: 15sevaluation_interval: 30sscrape_configs:- job_name: 'spring-apps'metrics_path: '/actuator/prometheus'scrape_interval: 10s # 对应用更频繁采集static_configs:- targets: ['app1:8080', 'app2:8080']relabel_configs:- source_labels: [__address__]target_label: instance- source_labels: [__meta_service_name]target_label: service

2. 关键优化参数

存储配置：

# 控制块存储行为
storage:tsdb:retention: 15d # 数据保留时间out_of_order_time_window: 1h # 允许乱序数据窗口# 限制内存使用
query:lookback-delta: 5mmax-concurrency: 20

四、Grafana 仪表盘设计

1. 核心监控仪表盘

JVM监控面板配置：

Panel 1: Heap Memory Usage
Query: sum(jvm_memory_used_bytes{area="heap"}) by (instance) / sum(jvm_memory_max_bytes{area="heap"}) by (instance)
Visualization: Time series with % unitPanel 2: GC Pause Time
Query: rate(jvm_gc_pause_seconds_sum[1m])
Visualization: HeatmapPanel 3: Thread States
Query: jvm_threads_states_threads{instance=~"$instance"}
Visualization: Stacked bar chart

2. 业务指标可视化

订单业务看板：

{"panels": [{"title": "Orders per Minute","targets": [{"expr": "rate(orders_total[1m])","legendFormat": "{{instance}}"}],"type": "graph","yaxes": [{"format": "ops"}]},{"title": "Processing Time (95%)","targets": [{"expr": "histogram_quantile(0.95, rate(orders_processing_time_seconds_bucket[1m]))","legendFormat": "P95"}],"type": "stat","unit": "s"}]
}

五、生产环境最佳实践

1. 指标命名规范

类型	前缀	示例
计数器	`_total`	`http_requests_total`
计量器	`_current`	`queue_size_current`
计时器	`_seconds`	`api_latency_seconds`
分布概要	`_summary`	`response_size_summary`

2. 标签使用原则

避免高基数标签：如用户ID、订单号等
统一标签命名：团队内保持一致（如env vs environment）
重要维度标记：region、az、service_version等

3. 资源优化技巧

Micrometer配置：

@Bean
MeterRegistryCustomizer<MeterRegistry> metricsCommonTags() {return registry -> registry.config().meterFilter(MeterFilter.deny(id -> {// 过滤不需要的指标return id.getName().startsWith("jvm_classes");})).commonTags("region", System.getenv("AWS_REGION"));
}

Prometheus资源限制：

# 容器部署时设置资源限制
resources:limits:memory: 8Girequests:cpu: 2memory: 4Gi

六、高级功能实现

1. 自定义Collector

public class CustomMetricsCollector extends Collector {@Overridepublic List<MetricFamilySamples> collect() {List<MetricFamilySamples> samples = new ArrayList<>();// 添加自定义指标samples.add(new MetricFamilySamples("custom_metric",Type.GAUGE,"Custom metric description",Collections.singletonList(new MetricFamilySamples.Sample("custom_metric",List.of("label1"),List.of("value1"),getCurrentValue()))));return samples;}
}// 注册Collector
new CustomMetricsCollector().register();

2. 告警规则示例

groups:
- name: application-alertsrules:- alert: HighErrorRateexpr: rate(http_server_requests_errors_total[5m]) / rate(http_server_requests_total[5m]) > 0.05for: 10mlabels:severity: criticalannotations:summary: "High error rate on {{ $labels.instance }}"description: "Error rate is {{ $value }}"- alert: GCTooLongexpr: rate(jvm_gc_pause_seconds_sum[1h]) > 0.1labels:severity: warning