如何通过Grafana监控基于spring-cloud-gcp-starter-pubsub的Google PubSub订阅指标？

Absolutely, there are reliable, battle-tested ways to track your Pub/Sub subscription's acknowledged message count and message latency in Grafana. Let’s break down the most practical approaches I’ve implemented or seen work well for teams using Spring Cloud GCP Pub/Sub:

1. Leverage Spring Boot Actuator + Micrometer (Simplest Native Integration) #

This is my go-to for Spring-based services since it’s baked into the ecosystem:

• Add Dependencies: Include spring-boot-starter-actuator and micrometer-registry-prometheus in your pom.xml or build.gradle—these let you expose metrics and make them scrapeable by Prometheus.
• Enable Pub/Sub Metrics: In your application.yml or application.properties, turn on Pub/Sub's built-in metrics:

  spring:
    cloud:
      gcp:
        pubsub:
          metrics:
            enabled: true
    management:
      endpoints:
        web:
          exposure:
            include: metrics,prometheus
      metrics:
        tags:
          application: your-service-name
  

• Key Metrics to Use:
  • Acknowledged Messages: Use pubsub_subscription_acknowledged_messages_total (a counter that tracks all successfully acknowledged messages per subscription).
  • Message Latency: Use pubsub_subscription_message_age_seconds (a gauge that shows the time between a message being published and being acknowledged, in seconds).
• Connect to Grafana: Set up Prometheus as a data source in Grafana, then build panels—use a Stat panel for total acknowledged messages, and a Time Series panel to visualize latency trends over time.

2. Use Google Cloud Monitoring (Native GCP Integration) #

If your service runs on GCP, this is a zero-code option for out-of-the-box metrics:

• GCP’s Built-in Pub/Sub Metrics: GCP automatically tracks metrics like:
  • Acknowledged messages (count of messages successfully acknowledged)
  • Message latency (time from message publication to acknowledgment)
• Connect Grafana to GCP Monitoring: Add the Google Cloud Monitoring data source in Grafana, authenticate with a service account that has the Monitoring Viewer role, then select the Pub/Sub metrics for your subscription directly in the panel builder.
• Pro Tip: You can filter metrics by subscription name, region, or project to narrow down to exactly what you need.

3. Custom Metrics (For Granular Control) #

If you need more specific tracking (e.g., latency per message type, or custom acknowledgment logic), roll your own with Micrometer:

• Track Acknowledged Messages: Create a Counter and increment it every time you successfully process a message:

  @Autowired
  private MeterRegistry meterRegistry;
  private Counter acknowledgedMessagesCounter;
  
  @PostConstruct
  public void initMetrics() {
    acknowledgedMessagesCounter = Counter.builder("pubsub.custom.acknowledged.messages")
      .tag("subscription", "your-subscription-name")
      .description("Total custom acknowledged messages")
      .register(meterRegistry);
  }
  
  // In your message listener method
  public void handleMessage(PubsubMessage message) {
    // Process message logic
    acknowledgedMessagesCounter.increment();
  }
  

• Track Custom Latency: Extract the message's publish time (either from PubSub's publish_time attribute or a custom attribute you add when sending), calculate the latency, and record it with a Timer:

  private Timer messageLatencyTimer;
  
  @PostConstruct
  public void initMetrics() {
    messageLatencyTimer = Timer.builder("pubsub.custom.message.latency")
      .tag("subscription", "your-subscription-name")
      .description("Latency from publish to acknowledgment")
      .register(meterRegistry);
  }
  
  public void handleMessage(PubsubMessage message) {
    long publishTime = Long.parseLong(message.getAttributes().get("publish_time"));
    long latencyMs = System.currentTimeMillis() - publishTime;
    messageLatencyTimer.record(latencyMs, TimeUnit.MILLISECONDS);
    // Process message logic
  }
  

• Expose & Visualize: These custom metrics will be included in the Prometheus endpoint—just add them to your Grafana panels like any other metric.

Quick Notes to Avoid Pitfalls #

• Make sure your Spring Cloud GCP version is 2.0 or higher (earlier versions don’t have full Micrometer integration for Pub/Sub).
• For latency metrics, clarify whether you need "time to receive" or "time to acknowledge"—the built-in message_age_seconds tracks the latter, which is usually what teams care about for processing SLAs.
• When using GCP Monitoring, ensure your service account has the correct IAM permissions to read metrics.

内容的提问来源于stack exchange，提问作者Mahmoud Azmy