Overview
This smaller section asks you to keep systems fast and affordable. Performance work covers monitoring application performance and acting on Active Assist insights and recommendations, which flag idle resources, oversized VMs and other improvements.
The FinOps subsection is broader. It includes controlling the cost of logging and monitoring, running fault-tolerant work on Spot VMs, rightsizing resources, planning spend with committed use discounts, sustained use discounts and network service tiers, using recommenders for cost, security, performance, manageability and reliability, and optimizing individual workloads on GKE, Cloud Run and Compute Engine. Questions usually ask for the saving that keeps reliability targets intact.
How it's tested
Cost and performance questions give a workload profile, such as steady, bursty, interruptible or latency-sensitive, plus a budget concern. You match the profile to the right pricing model, recommender action or configuration change. Read for words like fault tolerant, predictable, idle or overprovisioned.
-
Match workload to pricing
You choose Spot VMs, committed use discounts, sustained use discounts or on-demand pricing based on interruption tolerance and usage predictability.
-
Act on a recommendation
Given Active Assist or recommender output, you decide whether to rightsize, delete idle resources or change a configuration.
-
Cut telemetry spend
You reduce logging and monitoring cost through exclusions, sampling and retention without losing required signals.
A nightly batch job on Compute Engine runs for four hours, checkpoints its progress and can restart if interrupted. The team pays on-demand prices. Because the job tolerates preemption, Spot VMs offer the largest discount. A committed use discount would not fit as well, since the VMs run only a few hours each day and the commitment is billed for the full term.
Objectives 2
Official objectives from the Google Cloud outline, in outline order.
5.1 Collecting performance information in Google CloudApplication performance monitoring and Active Assist.
- What it means
- This covers monitoring application performance and using Active Assist insights and recommendations to find performance and efficiency improvements.
- What to understand
- Application performance monitoring on Google Cloud uses Cloud Trace for request latency, Cloud Profiler for CPU and memory use inside code, and Cloud Monitoring for latency and saturation metrics. Active Assist is the portfolio of recommenders and insights that analyze usage, such as VM machine type recommendations, idle resource detection and GKE cost suggestions. Recommendations can be viewed in the Recommendation Hub, applied, or exported for review.
- How it shows up
- Expect a performance or efficiency gap where you choose Profiler, Trace or a specific recommender.
- Common confusion
- Cloud Profiler examines code inside one service; Cloud Trace measures time across service calls.
5.2 Implementing FinOps practices for optimizing resource utilization and costsObservability costs, Spot VMs, rightsizing, discounts, network tiers, recommenders and workload tuning.
- What it means
- This covers managing observability costs, using Spot VMs, optimizing resource use, planning infrastructure costs with committed use discounts, sustained use discounts and network tiers, using Google Cloud recommenders for cost, security, performance, manageability and reliability, and optimizing costs for individual workloads on GKE, Cloud Run and Compute Engine.
- What to understand
- Spot VMs give deep discounts but can be stopped at any time, so they suit stateless or checkpointed jobs. Committed use discounts trade a one- or three-year commitment for lower rates on steady usage; sustained use discounts apply automatically to eligible VMs running much of the month. Premium Tier routes traffic over Google's network, while Standard Tier is cheaper and uses the public internet sooner. On GKE, rightsizing requests, cluster autoscaler and Autopilot reduce waste; Cloud Run can scale to zero.
- How it shows up
- A workload and spending pattern are described, and you choose the discount, tier or tuning that cuts cost safely.
- Common confusion
- Reservations guarantee capacity; committed use discounts lower price. They can be combined but are not the same.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Active Assist
Google's family of recommenders and insights.
- Remember
- Covers cost, security, performance, manageability and reliability.
- Tested as
- Optimization scenarios.
- Trap
- Building custom scripts to find idle VMs.
-
Machine type recommender
Suggests rightsizing VMs based on observed usage.
- Remember
- Uses recent utilization history.
- Tested as
- Overprovisioned VM questions.
- Trap
- Resizing without data.
-
Idle resource recommendations
Finds idle VMs, disks, IP addresses and similar resources.
- Remember
- Delete or stop what is unused.
- Tested as
- Waste reduction.
- Trap
- Paying for unattached persistent disks.
-
Spot VMs
Deeply discounted VMs that can be preempted.
- Remember
- Use for fault-tolerant, restartable work.
- Tested as
- Batch and CI workloads.
- Trap
- Running stateful databases on them.
-
Committed use discounts
Lower prices for a one- or three-year commitment.
- Remember
- Best for steady baseline usage.
- Tested as
- Predictable workloads.
- Trap
- Committing for short-lived or bursty use.
-
Sustained use discounts
Automatic discounts for eligible VMs running much of a month.
- Remember
- No action or commitment needed.
- Tested as
- Pricing comparison questions.
- Trap
- Thinking they require a purchase.
-
Network service tiers
Premium uses Google's backbone; Standard uses the public internet sooner.
- Remember
- Standard costs less but has lower performance.
- Tested as
- Egress cost decisions.
- Trap
- Choosing Standard for global latency-sensitive apps.
-
Observability cost
Charges for log ingestion, retention and metric volume.
- Remember
- Exclusions, sampling and retention tuning reduce it.
- Tested as
- Telemetry cost questions.
- Trap
- Dropping logs needed for audits.
-
GKE cost optimization
Rightsizing requests, autoscaling and Autopilot.
- Remember
- Pay for requested Pod resources in Autopilot.
- Tested as
- Container cost questions.
- Trap
- Leaving requests far above real use.
-
Cloud Run scale to zero
Instances drop to zero when there is no traffic.
- Remember
- Minimum instances trade cost for cold start latency.
- Tested as
- Serverless cost tuning.
- Trap
- Setting high minimum instances by default.
-
Cloud Profiler
Continuous low-overhead profiling of CPU and memory in code.
- Remember
- Finds expensive functions.
- Tested as
- Performance tuning.
- Trap
- Confusing it with tracing.
-
Billing export to BigQuery
Detailed cost data for analysis and attribution.
- Remember
- Use labels to attribute spend to teams.
- Tested as
- FinOps reporting.
- Trap
- Relying only on the console summary.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Compute pricing models
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| On-demand | Pay per second with no commitment. | Full flexibility at list price. | Paying it for steady 24x7 workloads. |
| Spot VMs | Large discount, can be preempted at any time. | Cheapest for interruptible work. | Using it for services that cannot restart. |
| Committed use discounts | Discount for a one- or three-year commitment. | Rewards predictable baseline usage. | You pay for the commitment even if usage falls. |
| Sustained use discounts | Automatic discount for long monthly runtime on eligible VMs. | No purchase required. | Not all machine families are eligible. |
Performance tools
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Cloud Trace | Latency of requests across services. | Shows where time is spent between calls. | It does not show which function uses CPU. |
| Cloud Profiler | CPU and heap profiles of running code. | Shows expensive functions inside a service. | It does not trace requests across services. |
| Active Assist recommenders | Usage-based improvement suggestions. | Finds idle and oversized resources automatically. | Recommendations need review before applying. |
Capacity versus price
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Reservation | Holds VM capacity in a zone. | Guarantees availability. | It does not lower price by itself. |
| Committed use discount | Lowers price for committed usage. | Reduces cost. | It does not guarantee capacity unless paired with a reservation. |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Spot for everything
- What candidates pick
- Move the production web tier to Spot VMs.
- Why it looks right
- Spot VMs are the cheapest option.
- Why it's wrong
- Preemption would cause outages and burn the error budget; Spot suits only interruptible work.
Clue to watch for The workload must be always available.
-
Commit to bursty use
- What candidates pick
- Purchase a three-year commitment for a seasonal workload.
- Why it looks right
- Commitments give large discounts.
- Why it's wrong
- The commitment is billed even in quiet months, so savings disappear for spiky or short-lived usage.
Clue to watch for Usage varies widely or the project ends soon.
-
Standard Tier for global users
- What candidates pick
- Switch a worldwide API to Standard Tier networking.
- Why it looks right
- Standard Tier is cheaper.
- Why it's wrong
- Global, latency-sensitive traffic benefits from Premium Tier routing over Google's network.
Clue to watch for Users spread across continents with latency targets.
-
Profiler for slow requests between services
- What candidates pick
- Enable Cloud Profiler to find a slow downstream call.
- Why it looks right
- Profiler is a performance tool.
- Why it's wrong
- Profiler examines code inside one process; Cloud Trace shows latency between services.
Clue to watch for Latency spread across several microservices.
-
Delete all debug logs
- What candidates pick
- Turn off logging to cut costs.
- Why it looks right
- Logging is a visible line item.
- Why it's wrong
- Removing all logs blinds troubleshooting; targeted exclusions, sampling and shorter retention save money safely.
Clue to watch for A cost request that still needs troubleshooting ability.
-
Manual rightsizing
- What candidates pick
- Review every VM by hand each quarter.
- Why it looks right
- It feels thorough.
- Why it's wrong
- Machine type and idle resource recommenders already analyze usage and suggest changes continuously.
Clue to watch for A large fleet and limited staff time.
Answer framework
Follow these steps to answer Section 5 questions about speed, waste or spending.
- Characterize the workload: steady or bursty, interruptible or critical, and how long it will run, before looking at any pricing option.
- Match the profile to pricing: Spot VMs for interruptible jobs, committed use discounts for steady baselines, sustained use discounts as an automatic bonus.
- Check Active Assist and recommenders for idle resources, oversized machines and GKE request changes before designing custom analysis.
- Tune per-workload settings such as GKE requests and autoscaling, Cloud Run minimum instances and network tier choice against latency needs.
- Confirm that the saving keeps SLOs and required telemetry intact, and track results with billing export and labels.
Practice question
Want more like this? The GCP PCDOE practice test has questions for every domain.
Checklist
0 of 10
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Because this section is about 12% of the exam, plan a focused block rather than weeks of study. Start by building a one-page table of pricing models: on-demand, Spot VMs, committed use discounts and sustained use discounts, with the workload profile each suits and its main risk. Add the two network tiers and the difference between a reservation and a commitment.
Then open the Recommendation Hub in a project with some running resources and read the machine type, idle resource and GKE cost recommendations. Enable Cloud Profiler on a sample service and compare its output with a Cloud Trace view. Review how logging exclusions and retention change observability cost, linking back to Section 4. Finish with practice questions that pair a workload description with a cost goal, and check that your answer never breaks a reliability requirement. If two options save similar amounts, prefer the one that needs less ongoing manual effort.
FAQ
How much of the GCP PCDOE exam is Optimizing performance and cost?
Google weights Section 5, Optimizing performance and cost, at about 12%, the smallest share. That is roughly 6 to 7 questions, as an editorial estimate.
What does FinOps mean in this exam?
It means treating cloud cost as a shared engineering practice: measuring spend, attributing it to teams and workloads, and using discounts, rightsizing and recommenders to improve value.
Do I need to memorize discount percentages?
No. Know which workloads each pricing model suits and its trade-offs. Exact percentages change and are unlikely to be tested.
What is the difference between Spot VMs and preemptible VMs?
Spot VMs are the current form of preemptible capacity and have no fixed 24-hour maximum runtime. Both can be stopped at short notice, so workloads must tolerate interruption.
Which recommenders should I know?
Know machine type rightsizing, idle resource recommendations, committed use discount recommendations and GKE cost suggestions, plus the fact that recommenders also cover security, performance, manageability and reliability.
Is GKE Autopilot cheaper than Standard?
It can be for clusters with poorly used nodes, since Autopilot bills for Pod requests rather than whole nodes. Well-packed Standard clusters with commitments can still cost less.
Other DevOps & Containers certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 5 knowledge
Check that you can match workloads to Spot VMs, commitments and recommenders while keeping reliability targets intact.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the GCP PCDOE AI tutor Get any concept explained your way