Metricbeat GCP模块采集BigQuery等GCP服务指标异常问题咨询

BigQuery指标采集报错 #

运行时抛出如下错误：

2021-12-06T09:51:10.206Z        ERROR   [gcp.metrics]   metrics/metricset.go:294        bigquery.googleapis.com/storage/table_count" metric descriptor is empty, this metric will not be collected

原有配置如下：

- module: gcp
  region: "us-central1"
  metricsets:
    - metrics
  project_id: "ProjectName"
  credentials_file_path: "/root/xyz.json"
  exclude_labels: false
  period: 1h
  metrics:
    - aligner: ALIGN_NONE
      service: bigquery
      service_metric_prefix: bigquery.googleapis.com/
      metric_types:
        - "bigquery.googleapis.com/storage/table_count"

Compute/Billing指标有字段无数据 #

Kibana发现页已显示对应字段名，但无实际数据返回，原有配置如下：

- module: gcp
  region: "us-central1"
  metricsets:
    - metrics
  project_id: "projectname"
  credentials_file_path: "/root/xyz.json"
  exclude_labels: false
  period: 1
  metrics:
    - aligner: ALIGN_NONE
      service: compute
      metric_types:
        - "instance/cpu/reserved_cores"
        - "instance/cpu/usage_time"
        - "instance/cpu/utilization"
        - "instance/uptime"

BigQuery配置修改 #

报错核心原因是service_metric_prefix配置和metric_types的全路径重复拼接，导致无法匹配到正确的指标描述符。调整后配置如下：

- module: gcp
  region: "us-central1"
  metricsets:
    - metrics
  project_id: "ProjectName"
  credentials_file_path: "/root/xyz.json"
  exclude_labels: false
  period: 3h # BigQuery存储类指标上报延迟高，建议拉长采集周期
  metrics:
    - aligner: ALIGN_MEAN # GAUGE类型指标用均值对齐更合理
      service: bigquery
      metric_types:
        - "storage/table_count" # 无需加前缀，service为bigquery时会自动补全官方前缀

Compute配置修改 #

无数据核心原因是period配置未加单位，默认按1秒处理，远小于GCP Compute指标最小60秒的上报粒度，无法拉取到数据。调整后配置如下：

- module: gcp
  region: "us-central1"
  metricsets:
    - metrics
  project_id: "projectname"
  credentials_file_path: "/root/xyz.json"
  exclude_labels: false
  period: 1m # 最低设为60秒，和Compute指标上报粒度匹配
  metrics:
    - aligner: ALIGN_MEAN
      service: compute
      metric_types:
        - "instance/cpu/reserved_cores"
        - "instance/cpu/usage_time"
        - "instance/cpu/utilization"
        - "instance/uptime"

通用检查项 #

• 确认GCP项目已启用Cloud Monitoring API
• 确认服务账号已绑定roles/monitoring.viewer权限，BigQuery采集还需额外授予bigquery.tables.list权限
• Billing指标需先在GCP控制台开启账单导出到Cloud Monitoring，默认不开启无法采集
• 首次配置完成后等待5-10分钟再到Kibana查询，GCP指标上报存在天然延迟

内容的提问来源于stack exchange，提问作者deev