Professional Cloud DevOps Engineer

Collecting performance information in Google Cloud

To optimize performance and cost, you first need to collect the right data about how your systems are running. Cloud Monitoring is the primary service for gathering metrics, logs, and metadata from your Google Cloud resources. You set up metric scopes to define which projects, folders, or organizations you want to monitor, and then you create dashboards and alerting policies to visualize and respond to the data. The key decision is choosing which metrics matter most: infrastructure metrics like CPU and memory usage tell you about your VMs and containers, while application metrics like request latency and error rates tell you about user experience. You typically want both, because infrastructure metrics can show you a resource is busy while application metrics reveal whether that busyness is actually hurting users.

Cloud Logging complements Monitoring by capturing detailed log entries from your services, including audit logs for security events and agent logs from the Cloud Logging agent running on your VMs. The relationship between these two services is important: you can create log-based metrics from Logging data and then use those metrics in Monitoring dashboards and alerts. For example, you might count the number of 500 errors in your application logs, turn that count into a metric, and then set an alert that fires when the error rate exceeds a threshold. This flow lets you monitor things that aren't captured by standard infrastructure metrics alone.

Cloud Trace helps you understand latency by collecting trace spans from your applications as requests travel through distributed services. Each span records how long a single operation takes, and traces connect spans together to show the full path of a request. You use this to find bottlenecks—the slowest service in a chain that makes everything else wait. Cloud Profiler takes a different approach by continuously sampling CPU and heap usage across your running applications, showing you which functions or methods consume the most resources. While Trace tells you where time is spent, Profiler tells you where compute resources are actually being used, which helps you target your optimization efforts to the code that matters most.

Implementing FinOps practices for optimizing resource utilization and costs

FinOps is the practice of bringing financial accountability to cloud spending, and in Google Cloud it relies on Cloud Billing data combined with Cloud Asset Inventory and Recommender. You start by setting up budgets and budget alerts in Cloud Billing to get notified when spending approaches or exceeds your limits. The Recommender service then analyzes your usage patterns and generates recommendations for rightsizing VMs, removing idle resources, and committing to committed use discounts (CUDs) for predictable workloads. The key workflow is: collect billing data, identify waste through recommendations, take action to reduce waste, and then monitor the results to see if your changes actually saved money.

Resource hierarchies play a critical role in cost management. You organize your projects into folders and then under an organization node, which lets you apply billing account permissions and budgets at different levels. For example, you might set a budget for a development folder that is separate from your production folder, and then use labels and tags to track costs by team, environment, or application. This structure lets you answer questions like "how much did the mobile app team spend in staging this month?" without manually sorting through invoices.

Committed use discounts and sustained use discounts are two ways to reduce costs for predictable workloads. Sustained use discounts apply automatically when you run a VM for a significant portion of a month, while committed use discounts require you to commit to spending a certain amount per month for one or three years in exchange for larger discounts. The tradeoff is flexibility versus savings: committed use discounts give you the best price but lock you into a spending level, so you should only commit to workloads that are stable and unlikely to change. Preemptible VMs and spot VMs offer even larger discounts for workloads that can tolerate interruptions, such as batch processing or fault-tolerant applications, but they can be terminated at any time when Google Cloud needs the capacity back.