cost-management

Grafana Cloud 成本管理

文档：https://grafana.com/docs/grafana-cloud/cost-management-and-billing/

通过自适应信号和成本归属标签，降低指标、日志和追踪的开销。

前提条件

• 已启用 Adaptive Metrics / Logs / Traces 的 Grafana Cloud 堆栈（可在 Cost Management 下查看）
• Alloy（或 Grafana Agent）正在采集遥测数据，且 API 密钥具有 metrics:write + logs:write（以及 traces:write）范围
• 对该堆栈拥有管理员权限，以应用 Adaptive 建议

常见工作流

1. 将成本归属到团队 / 服务

# 1. Add external labels in Alloy (metrics + logs configs)
prometheus.remote_write "cloud" {
  endpoint { url = sys.env("PROMETHEUS_URL") /* ... */ }
  external_labels = { team = "platform", project = "checkout-service" }
}

# 2. Reload Alloy
curl -X POST http://localhost:12345/-/reload

# 3. Verify labels arrived in Grafana Cloud
#    In Explore, run:  count by (team, project) ({__name__=~".+"})
#    Then visit Cost Management → group by `team` / `project`

完整的 Alloy 代码片段请参见 references/adaptive-signals.md。

2. 使用 Adaptive Metrics 降低指标基数

# 1. Pull recommendations
curl https://<stack>.grafana.net/api/plugins/grafana-adaptive-metrics-app/resources/v1/recommendations \
  -H "Authorization: Bearer <token>" | jq '.recommendations | length'

# 2. In the UI: Grafana Cloud → Adaptive Metrics → review rules sorted by series-reduction impact
# 3. Test in "Preview" mode before applying
# 4. Apply (takes effect within 5 min)

# 5. Verify — series count should drop on the affected metrics
#    Before applying, capture baseline:
#      count({__name__="http_request_duration_seconds_bucket"})
#    Wait 10 min after apply, run again — expect 10x+ reduction for high-card metrics.

# Rollback if needed: open the rule in the UI → Disable, or DELETE /v1/rules/<id>.

3. 在 Alloy 中丢弃嘈杂的日志

# 1. Add a filter stage (see references/adaptive-signals.md for the full block)
loki.process "filter_logs" {
  forward_to = [loki.write.cloud.receiver]
  stage.drop { expression = ".*GET /health.*" }
}

# 2. Reload Alloy
curl -X POST http://localhost:12345/-/reload

# 3. Verify the filter — health logs should NOT appear in Logs Drilldown
#    LogQL check (should return 0):
#      sum(rate({app="my-app"} |= "GET /health" [5m]))
#    Bytes-ingested should also drop. Compare 24h before/after:
#      sum(increase(loki_ingester_chunk_size_bytes_sum[24h])) by (namespace)

4. 在达到配额前设置用量告警

可在 references/alerts-and-queries.md 查看可直接粘贴的规则（MetricsUsageHigh、LogsIngestionHigh）。

优化清单

• 应用 Adaptive Metrics 建议 — 通常可减少 40-60% 的序列
• 在 Alloy 中丢弃健康检查 / 就绪探针日志
• 对追踪进行尾部采样，采样率设为 5-10%，并保留错误 / 慢速跨度
• 在每个 Alloy 配置中添加 team + project 外部标签
• 在达到配额的 80% 时设置用量告警
• 用录制规则替代开销高的临时查询

参考资料

• references/adaptive-signals.md — Adaptive Metrics / Logs / Traces 配置；成本归属标签
• references/alerts-and-queries.md — 用量告警规则、用于查找成本的 PromQL、计费单位表

资源

• 成本管理文档
• Adaptive Metrics
• Adaptive Logs