Observability Architecture

Exam guide§3.4

How the pieces fit together before you configure any of them. Monitoring and Logging each follow a layered architecture, and how you collect application metrics depends on where the workload runs.

Cloud Monitoring architecture

A Cloud Monitoring deployment has three layers - signals flow from metric collection, through metric storage (the Cloud Monitoring API), to visualization and analysis:

The three layers of a Cloud Monitoring deployment: collection, storage (Cloud Monitoring API), and visualization & analysis - plus hybrid/multi-cloud inputs and third-party export (Grafana) and notifications.

Platform monitoring (default, free)

Platform monitoring is blackbox visibility into your Google Cloud services, and it is on by default - system metrics are collected with no effort on your part.

Platform monitoring: Google Cloud services (100+ services, GKE, Compute Engine) automatically ship system metrics into Cloud Monitoring, which powers dashboards, uptime checks, and alerting - no agents or setup required.

#NumbersSystem metrics

• Over 1,500 metrics across more than 100 Google Cloud services, at no cost.
• Compute Engine alone reports 25+ metrics per VM instance.
• To pull these into a third-party monitoring product, ingest them through the Cloud Monitoring API.

Application monitoring on GKE

For workloads on GKE, use a Prometheus-based approach via Google Managed Service for Prometheus (GMP) - part of Cloud Monitoring. It ingests data in Prometheus format, supports PromQL, and includes the Prometheus expression browser and rule evaluation, so GKE cluster and workload metrics show up as Prometheus data.

On GKE, Prometheus-based collectors plus the free GKE and control-plane metrics feed Cloud Monitoring, which powers dashboards, uptime checks, alerting, and notifications. Google Managed Service for Prometheus is the recommended path.

Application monitoring on Compute Engine

For Compute Engine VMs, the Ops Agent collects in-process and third-party application metrics (30+ plugins, built on OpenTelemetry) and ships them through the Cloud Monitoring API.

Ops Agent collection: Compute Engine plus the Ops Agent ship host, process, and third-party metrics through the Cloud Monitoring API into dashboards, uptime checks, alerts, and notifications (and out to Grafana).

The Cloud Monitoring page has the same path drawn as a hand-authored diagram, plus the guest-metrics gotcha (memory and disk space need the agent).

Hybrid and multi-cloud

To bring metrics and logs from on-premises VMs and other clouds (AWS, Azure, and others) into Google Cloud, use Blue Medora's BindPlane. BindPlane metrics arrive as custom metrics (chargeable), and its logs are billed like other Cloud Logging logs.

Hybrid and multi-cloud: BindPlane by Blue Medora imports monitoring and logging data from on-premises and other clouds into Cloud Monitoring and Cloud Logging for one pane of glass.

The Partner integrations page has the full BindPlane reference architecture as a hand-authored diagram.