3.4 - Operational tooling: remediation, observability and safe releases

1. An S3 bucket stores logs that are read a lot for 30 days, occasionally for a year, then must be kept 7 years for compliance but almost never read. What minimizes storage cost automatically?

• Keep everything in S3 Standard the whole time and delete after 7 years
• A lifecycle policy: Standard -> Standard-IA -> Glacier Deep Archive over time (correct answer)
• S3 Intelligent-Tiering enabled on the bucket with no lifecycle rules at all
• Cross-Region replication of every object to a slightly cheaper Region

A lifecycle policy transitions objects to cheaper tiers as access drops (Standard for 30 days, Standard-IA for the year, Glacier Deep Archive for the 7-year cold retention), then expires them. Standard-only wastes money; Intelligent-Tiering helps unknown patterns but Deep Archive is cheaper for known cold data; replication adds cost.

2. A finance team asks which running services drive most of the monthly bill and wants to slice the spend by team and account. Which tool answers this fastest?

• Parse the raw CloudTrail management event history and attribute each charge to a team by hand
• AWS Cost Explorer, filtered and grouped by service, linked account, and cost allocation tag (correct answer)
• Build CloudWatch metric-math dashboards that estimate the per-service monthly bill from usage metrics
• Enable AWS Config recorders and read each resource configuration timeline to infer its cost

Cost Explorer natively groups spend by service, linked account, and activated cost allocation tags, giving the breakdown in seconds. CloudTrail and Config record API activity and configuration, not dollar amounts, and CloudWatch metric math does not know unit prices.

3. After a deployment error left resources running, the company wants to be alerted automatically the next time spend deviates sharply from normal patterns, without setting fixed dollar thresholds. What should they enable?

• A scheduled Lambda that queries the Cost Explorer API nightly and diffs totals against yesterday
• An AWS Budgets fixed monthly threshold that emails the team when the amount is exceeded
• AWS Cost Anomaly Detection with a monitor and an alert subscription (correct answer)
• A CloudWatch billing alarm on the EstimatedCharges metric with a static threshold

Cost Anomaly Detection uses machine learning to learn normal spend and flags deviations without any manual threshold, notifying via an alert subscription. Budgets and billing alarms rely on static thresholds you must guess, and a custom Lambda reinvents what the managed service already does.

4. A steady EC2 fleet changes instance families and moves between Regions as teams refactor, and even shifts some load to Fargate and Lambda. The company wants a commitment discount that keeps applying through all these changes. Which purchase fits best?

• Compute Savings Plans committing to an hourly dollar amount (correct answer)
• Standard Reserved Instances scoped to one instance family in one Region
• EC2 Instance Savings Plans locked to a single instance family per Region
• On-Demand Capacity Reservations renewed manually each month

Compute Savings Plans apply to any instance family, size, Region, OS, tenancy, and even Fargate and Lambda, so the discount follows the workload as it moves. EC2 Instance Savings Plans and Standard RIs lock to a family/Region, and Capacity Reservations guarantee capacity but give no commitment discount by themselves.

5. A team already runs on Reserved Instances but expects to switch instance families as they adopt newer generations over the three-year term, and wants to preserve the reservation discount when they do. Which reservation type should they hold?

• Standard Reserved Instances, which give the deepest discount but cannot change family
• On-Demand pricing kept until the exact future family is finally decided
• Convertible Reserved Instances, which can be exchanged for a different family (correct answer)
• Spot Instances requested against the newest generation each quarter

Convertible RIs can be exchanged for RIs of a different instance family, OS, or tenancy of equal or greater value, preserving a commitment discount across generation changes. Standard RIs give a bigger discount but cannot change family, On-Demand carries no discount, and Spot is interruptible and unsuited to steady reserved capacity.

6. After buying Savings Plans, a company sees that a large share of eligible On-Demand usage is still billed at full price. Which metric should they watch to decide whether to buy more commitment?

• The Savings Plans utilization percentage, which measures how much of the commitment goes unused
• The reserved instance modification count reported in the billing console each cycle
• The number of active Compute Optimizer findings across the running instance fleet
• Savings Plans coverage, the share of eligible usage covered by a plan (correct answer)

Coverage shows the percentage of eligible spend that a Savings Plan is discounting, so low coverage means room to commit more. Utilization instead shows whether an existing commitment is being fully consumed, and Compute Optimizer and RI modification counts do not measure commitment gaps.