Observability with Prometheus and Grafana

Google Cloud Managed Service for Prometheus

Google Cloud’s Managed Service for Prometheus (GMP) is a fully managed, Prometheus-compatible monitoring service on Google Cloud. It uses Google’s Monarch time-series database (the same backend that Google uses internally) to achieve global scalability. GMP lets you continue using PromQL and existing Prometheus exporters, while offloading storage and scaling to Google Cloud.

• Free Tier & Pricing: Google Managed Service for Prometheus does not have a separate free tier for Prometheus-style metric ingestion. Every custom metric sample is billed from the first sample. Pricing starts at $0.06 per million samples for the first 50 billion and decreases in tiers for higher volumes. Storage and long-term retention are included at no extra cost. Read API usage is free for the first 1 million time series read per month, then costs $0.50 per million. Only Google’s built-in Cloud Monitoring metrics are free to query with PromQL; GMP ingestion itself has no free allotment.

• Storage Retention: All metrics are retained for 24 months (2 years) by default at no extra cost. The service stores full-resolution data for 1 week, then automatically down-samples older data (1-minute resolution for data 1 week-6 weeks old, and 10-minute resolution beyond 6 weeks, up to 24 months). This long retention (with automated downsampling) means you can analyze up to two years of metrics for trends without managing any storage. There are no limits on total time-series count or cardinality - GMP can handle very large metric volumes since data is stored in Monarch.

ℹ️ Downsampling is the process of reducing the resolution of time-series data over time to save storage space. For example, if you initially started pulling metrics at an interval of 15 seconds, downsampling to 1-minute intervals means that older data points are produced by averaging or aggregating the original 15-second samples into 1-minute buckets.

Here is a simple example of downsampling over time in general:

1) First step: Collect high-resolution data (every 15 seconds):

00:00:00 - 70%
00:00:15 - 72%
00:00:30 - 69%
00:00:45 - 74%
--
00:01:00 - 71%
00:01:15 - 73%
00:01:30 - 68%
00:01:45 - 75%

2) Second step: Downsample to 1-minute resolution after 1 week:

00:00:00 - 71