使用Java SDK向Google Stackdriver推送自定义指标时出错

Hey there! Let's walk through the common issues that pop up when pushing custom metrics to Google Stackdriver with the Java SDK—this stuff can be tricky, but we'll break it down step by step to get you back on track.

• Make sure your service account has the right permissions: it needs monitoring.metricDescriptors.create (to define new custom metrics) and monitoring.timeSeries.create (to send data points). The easiest way is to assign the Monitoring Metric Writer IAM role to your service account, or use more granular permissions if you need tighter control.
• Double-check that you're loading the correct service account key file in your code. Your initialization should look something like this (don't botch the file path or key content!):

GoogleCredentials credentials = GoogleCredentials.fromStream(new FileInputStream("/path/to/your-service-account-key.json"));
MonitoringServiceClient client = MonitoringServiceClient.create(MonitoringServiceSettings.newBuilder().setCredentialsProvider(FixedCredentialsProvider.create(credentials)).build());

Stackdriver is picky about metric formatting—here's what to check:

• Metric IDs can only use lowercase letters, numbers, and underscores, and can't start with a number. So my_app_api_latency works, but 123_requests will fail.
• The type field in your MetricDescriptor must follow the format custom.googleapis.com/[your_metric_name]. For example: custom.googleapis.com/app/backend_response_time
• Stick to Stackdriver's predefined units (like s for seconds, count for totals) instead of making up your own. Using invalid units will trigger an error.

• Use the latest version of the Google Cloud Monitoring Java SDK—old versions might have bugs or lack support for newer API features. If you're using Maven, update your dependency:

<dependency>
    <groupId>com.google.cloud</groupId>
    <artifactId>google-cloud-monitoring</artifactId>
    <version>最新版本号</version>
</dependency>

• Verify your TimeSeries construction:
  • Pick the right MonitoredResource type. For most custom metrics not tied to a specific GCP resource, global works perfectly. If you're targeting a GCE instance or GKE pod, use the corresponding resource type (like gce_instance).
  • Ensure your Point has a valid Interval—Stackdriver rejects data points that are too far in the past (usually older than a few hours). Set the start and end time to the current timestamp (or a small window around it).
  • Match the value type to your metric: use Int64Value for integer counters, DoubleValue for gauges/latency metrics, and avoid stuffing strings into numeric fields.

• Enable debug logging for the SDK to see exactly what Stackdriver is complaining about. In Java, you can set the log level to DEBUG for the com.google.cloud.monitoring package—this will spit out raw API responses, which often include specific messages like "Invalid metric type" or "Permission denied".
• Test with a direct API call using curl to rule out SDK issues. Here's a sample command (replace placeholders with your project ID and metric data):

curl -X POST -H "Authorization: Bearer $(gcloud auth application-default print-access-token)" \
-H "Content-Type: application/json" \
https://monitoring.googleapis.com/v3/projects/your-project-id/timeSeries \
-d @metric-data.json

If this succeeds, the problem is in your Java code; if it fails, you've got an issue with permissions or metric definition.

• Don't repeatedly create the same MetricDescriptor—Stackdriver will throw an "Already exists" error. Add a check to see if the metric exists before calling createMetricDescriptor.
• Watch out for quota limits: Stackdriver caps the number of data points you can send per second. If you're hitting throttling errors, batch your time series data or reduce your reporting frequency.

内容的提问来源于stack exchange，提问作者Adam Sandor