Prepare and test your skills

Prepare and test your skills

The Cloud Monitoring signals used by Gemini Cloud Assist (opens in a new tab) are health metrics, logs and traces, and alert and incident events collected from Google Cloud projects and applications. Gemini Cloud Assist analyzes these three signal types together to explain health, performance, and reliability issues. When metrics, logs, and events are correlated, unusual behavior can be detected and linked to a resource problem.
Health metrics analyzed by Gemini Cloud Assist are time-series measurements that show how resources perform over a selected time period. Gemini Cloud Assist reviews metrics such as utilization, latency, and errors for Google Cloud products, services, and workloads. Concrete examples include current CPU utilization of a database, total virtual machine (VM) network bytes sent per second for each instance, queries per second (QPS) on SQL databases, VM utilization, and estimated backlog processing time for Dataflow jobs. When a metric moves outside its normal pattern, Gemini Cloud Assist can flag it as a possible anomaly and connect it to slow queries, high database load, or overloaded infrastructure.
Logs that provide diagnostic context are detailed records of what happened inside resources, and Gemini Cloud Assist uses them to explain metric changes. Relevant log signals include database error log entries, Cloud SQL query insights data such as high-latency and slowest queries, and additional product observability data with logs and traces. Aggregation of logs, metrics, and traces for App Hub applications allows health views to show related evidence in one place. By linking an error spike or latency increase to specific log entries and query details, Gemini Cloud Assist helps show whether a resource issue comes from code, configuration, capacity, or failed operations.
Alert and incident events that signal problems are markers that flag important occurrences on health charts alongside metrics. Cloud Monitoring (opens in a new tab) charts work with three event types:
Charts show recommended events by default, and the Annotations selector controls which event types are displayed. Open alerts count policy-based alerts for a project or application, while Google Cloud incidents count platform-side outages viewed through Personalized Service Health. This separation lets Gemini Cloud Assist distinguish a problem in your configuration from a wider Google Cloud service problem.
Exam tip: Open alerts come from your own Cloud Monitoring alerting policies, while Google Cloud incidents come from Service Health outages or degradations.
How signals combine for observability is shown on the Health & troubleshooting page in Cloud Hub, which merges metrics, logs, and events into one view for a selected time range. The default time range is the last hour. Selecting a product, service, or workload shows its key health charts with annotated events, plus links to open alerts, Google Cloud incidents, detailed logs, and Monitoring dashboards.
Three monitoring signals — health metrics, logs and traces, and alert and incident events — flow through Cloud Monitoring into Gemini Cloud Assist, which correlates them on the Health and troubleshooting page to produce root-cause findings.
| View | Scope | What the list shows |
|---|---|---|
| Project view | Individual project | Open alerts and Service Health events for each active Google Cloud product, with key health charts below the product list |
| Application view | App Hub application | Open alerts and Service Health events for each service or workload in the application, with key health charts below the services list |
From either view, you can open View alerts to inspect policy-based incidents, View incidents to check platform outages, View details to review extra observability data and logs for a product, or View dashboard to open the workload dashboard in Monitoring. By correlating a metric anomaly with its alerts, related Service Health incidents, and supporting logs, Gemini Cloud Assist investigations can perform root-cause analysis (RCA) and guide the next troubleshooting steps.
Gemini Cloud Assist (opens in a new tab) insights and recommendations are AI-generated explanations of monitoring data and suggested actions for Google Cloud resources. Gemini Cloud Assist reads signals from Cloud Monitoring, recommendations from the Recommender API, and outage data from Service Health, then presents insights in the Cloud Assist panel and on the Health and troubleshooting page. An insight describes the current state of a monitored resource, while a recommendation suggests a change to improve performance, cost, or reliability.
Performance and reliability insights in Gemini Cloud Assist describe how a resource is behaving right now, using metrics such as utilization, latency, errors, and backlog. A high CPU utilization, slow query, high error-log count, or large Dataflow backlog means the underlying database, virtual machine (VM), or job is under stress or inefficient. You can ask about this state in plain language, and Gemini Cloud Assist returns the relevant metric or finding for that specific resource.
Common performance questions include:
What is the current CPU utilization of this database?Which of my queries have high latency for this database instance?What is the slowest query for this database in this instance?Which of my VMs have the highest utilization?What are the top 5 dataflow jobs with the highest estimated backlog processing time?These answers relate directly to the monitored state because they reflect live or recent telemetry from Cloud Monitoring and Query Insights. If Gemini Cloud Assist reports idle VMs, high network bytes sent, or low queries per second (QPS), it is summarizing the actual usage pattern of those VMs or Cloud SQL instances.
Cost and resource optimization recommendations in Gemini Cloud Assist identify wasted capacity and more efficient configurations. This help requires the Recommender API (recommender.googleapis.com) and the Recommender Viewer role, which let Gemini Cloud Assist surface estimated savings and reliability findings. Typical prompts are Give me recommendations to reduce my resource costs, Which of my VMs are idle, What BigQuery datasets should I partition or cluster, and What are all the quotas for my project that are almost hitting the limit.
An idle-VM finding means the VM exists and incurs cost but shows little CPU or network use, so stopping or rightsizing it would save money. A partition-or-cluster finding means a BigQuery dataset is being scanned inefficiently, so changing its layout would improve query speed and lower scanned bytes. A near-limit quota finding means usage is approaching the project limit, so the resource is at risk of throttling or failed deployments. Gemini Cloud Assist generally does not provide exact billing amounts because pricing depends on many factors, so billing details remain in the FinOps hub.
Security, compliance, and service health findings in Gemini Cloud Assist separate problems in your configuration from problems in Google Cloud itself. Database prompts such as What issues are affecting my databases and What databases have NIST issues in Europe reflect compliance and health state, such as misconfiguration, location, or failure to meet a standard. Storage questions answered through Storage Insights datasets, such as bytes stored in us-central1 or usage broken down by content type, reflect the actual inventory and placement of buckets and objects.
Service health findings come from a different source than your own alerts. Open alerts are firing alerting policies you created in Cloud Monitoring (opens in a new tab), which Monitoring calls incidents, while Google Cloud incidents are outages or degradations reported by Google through Service Health. Checking health requires the Service Health API and Service Health Viewer role, with prompts such as How is my project impacted by the Cloud Storage incident or Is there a workaround for the BigQuery incident.
How insights connect to monitored resource state is shown on the Health and troubleshooting page, which maps each insight to a product, service, or workload over a selected time range. The default time range is the last hour. The page lists open alerts and Google Cloud incidents, then shows per-product or per-service health charts with metrics such as utilization, latency, or errors. Charts also overlay events, including Cloud Monitoring alert events, Service Health events, and product-specific events, with recommended events shown by default.
| Insight type | What it tells you | Underlying resource state |
|---|---|---|
| Performance insight | Current speed, load, or errors | A VM, database, or job is busy, slow, or failing, shown in charts and logs |
| Cost recommendation | Wasted spend or inefficient layout | A resource is idle, oversized, or poorly partitioned, shown in usage metrics |
| Health finding | Your alert versus Google outage | Your alerting policy is firing, or a Google Cloud incident affects your project |
To move from insight to guided remediation, Gemini Cloud Assist investigations provide root-cause analysis for infrastructure and application issues. An investigation flow moves from chat to analysis in order: first you ask Gemini Cloud Assist to investigate an issue, then you click Create investigation, then you review and edit the pre-populated details, and only after you click Create does Gemini Cloud Assist run the investigation. Creating investigations requires a Premium Support contract and the Investigation Creator role.
Exam tip: Open alerts come from your own Cloud Monitoring alerting policies, while Google Cloud incidents come from Service Health and describe Google-side outages.
Guided remediation in Gemini Cloud Assist (opens in a new tab) is the workflow where AI analysis turns monitoring signals into specific, reviewable next steps for reliable operations. Gemini Cloud Assist ingests logs, metrics, configurations, and page context from the Google Cloud console, then proposes explanations, ranked hypotheses, and recommended actions instead of leaving the operator to interpret raw data alone. Adopting this workflow helps teams resolve issues faster, prevent repeats, and stay aligned with Google Cloud operational best practices.
How Gemini Cloud Assist proposes remediation steps involves two connected paths: interactive chat and Gemini Cloud Assist Investigations. In chat, the Cloud Assist panel (opens in a new tab) in the Google Cloud console answers prompts about an error, a slow service, or a best practice by explaining the message in plain terms, suggesting which metrics to check first, and listing troubleshooting steps and tools. In Investigations, the system performs deeper automated analysis that is integrated into workflows like Logs Explorer, Bigtable System insights, and Cloud Run, exploring multiple hypotheses in parallel across multi-tier topologies.
When an investigation runs from an error or a specific resource, data flows from logs, metrics, and configurations to the investigation engine, and the engine returns structured results to the Investigation details pane. The results are organized into three parts that build on each other:
For services deployed from source code in Cloud Run, investigations can also send code fix suggestions to the Source tab to help resolve build and deployment issues. This structure moves operations from reactive log searching to guided action, because each hypothesis is tied to observed evidence rather than a generic suggestion.
How to validate remediation before applying it means checking every AI-generated explanation and recommendation against trusted operational data before changing production. Gemini Cloud Assist is an early-stage technology that can generate output that seems plausible but is factually incorrect, so all output must be validated before use. Page context awareness, memory of session history, and links under Show related content help make responses contextual, but they do not replace operator review.
A practical validation workflow follows an ordered sequence:
In some analyses Gemini Cloud Assist detects a change but returns no recommendation, which is itself a signal to keep investigating rather than applying an unproven fix.
Exam tip: Always validate Gemini Cloud Assist output before use, because plausible-sounding explanations can be factually incorrect.
Operational best practices supported by guided workflows are repeatable habits that keep environments healthy and prevent the same failure from returning. Proactive validation is one habit: check configurations locally before pushing them, such as using nomos vet for syntax validation and nomos hydrate to preview rendered Kustomize or Helm configurations for Config Sync. Another habit is to separate platform-wide scope from team scope, such as using cluster-scoped controls for administrators and namespace-scoped controls for application teams, which reduces the affected area of a misconfiguration.
Guided operations also connect monitoring to the wider support and optimization lifecycle. If a support package is available, investigation results can be transferred to a Google Cloud support case to provide context that resolves the case faster, and chat can automatically pre-fill support case fields by summarizing environment states and logs. Teams can check for platform incidents with Service Health prompts, request cost and reliability recommendations through the Recommender API, and ask for best-practice guidance for securing a Google Kubernetes Engine (GKE) cluster or troubleshooting a NotReady node or CrashLoopBackOff Pod. Together, these workflows support reliable operations because detection, remediation, validation, and escalation share the same evidence and history.
Open alerts come from customer-configured Cloud Monitoring alerting policies that detect threshold breaches, while Google Cloud incidents come from Service Health outages or degradations. This separation lets Gemini Cloud Assist distinguish a problem in your configuration from a wider Google Cloud service problem.
Gemini Cloud Assist analyzes health metrics, logs and traces, and alert and incident events collected from Google Cloud projects and applications. It correlates these three signal types together to detect unusual behavior, flag possible anomalies, and link them to a resource problem.
Gemini Cloud Assist investigation results include an Issue description with start and stop time, ranked Findings, and Hypotheses for next remediation steps, and all output must be validated before changing production. Validation is required because Gemini Cloud Assist can generate plausible-sounding output that is factually incorrect.
An operations engineer is troubleshooting intermittent application crashes detected in Google Cloud Monitoring. To identify the root cause and restore system reliability, the engineer initiates an automated investigation using Gemini Cloud Assist from the Logs Explorer.
Gemini Cloud Assist evaluates logs, metrics, and resource configurations, producing ranked observations, potential root-cause hypotheses, and suggested remediation steps.
Following Google Cloud operational best practices, what should the engineer do to safely validate and apply the proposed remediation?
Review the source data links included with the observations to fact-check the findings, validate the suggested fixes against organizational standards, and apply the remediation.
Forward the investigation directly to Cloud Customer Care and avoid testing or applying any remediations internally until Google support replies.
Configure Gemini Cloud Assist to execute autonomous self-healing scripts immediately across all production clusters to prevent human delay.
Accept the highest-ranked hypothesis as guaranteed fact and immediately deploy the recommended changes without verifying the underlying telemetry.
Last updated