Re:Invent 2024 Interview Guide

Amazon EKS Auto Mode was announced at Re:Invent 2024 to deliver a fully managed Kubernetes experience:

What EKS Auto Manages For You

• Automatic Compute — No node groups to configure. EKS Auto provisions the right EC2 instances (powered by Karpenter) based on pod requirements. Supports Spot, On-Demand, and Graviton automatically.
• Automatic Networking — VPC CNI, CoreDNS, and kube-proxy are managed and upgraded automatically. Load balancer integration is built in.
• Automatic Storage — EBS CSI driver is pre-installed. PersistentVolumeClaims just work.
• Automatic Upgrades — K8s version, AMIs, and add-ons are upgraded automatically with minimal disruption.

EKS Auto vs Standard EKS

Amazon Aurora DSQL (Distributed SQL) — the first serverless, multi-region distributed SQL database on AWS:

• Distributed & Multi-Region — Active-active writes across multiple AWS regions with strong consistency.
• PostgreSQL Compatible — Standard PostgreSQL interface, existing apps work without modification.
• Serverless — No capacity planning, no instances. Scales automatically.
• Serializable Isolation — Unlike DynamoDB Global Tables (eventual consistency), DSQL provides serializable isolation across regions.

Aurora DSQL vs Alternatives

Amazon S3 Tables — native Apache Iceberg table support in S3, announced at Re:Invent 2024:

• Structured Data in S3 — Store tabular data directly in S3 with built-in table format support. No separate metastore needed.
• Apache Iceberg Format — Native Iceberg support with schema evolution, time travel, partition evolution, and ACID transactions.
• Query with Standard Tools — Query from Athena, EMR, Redshift, and any Iceberg-compatible tool.
• Auto-Optimization — S3 manages compaction, snapshot management, and orphaned file cleanup automatically.

S3 Tables vs Traditional Lakehouse

Interviewers ask about recent AWS launches to test if you stay current. Knowing these services and when to use them signals a growth mindset.

1. Explain EKS Auto Mode. How does it differ from standard EKS? When would you NOT want to use it?
   Answer Guide

   EKS Auto Mode manages compute, networking, and storage automatically (no node groups, no Karpenter config, no add-ons). NOT ideal when: you need custom AMIs, specific kernel configurations, GPU workloads with custom drivers, or fine-grained node control. Trade-off: simplicity vs control.
2. Compare Aurora DSQL with DynamoDB Global Tables and Aurora Global Database. When would you choose DSQL over the other two?
   Answer Guide

   DSQL = distributed SQL with strong consistency across regions (fills the gap between Aurora Global DB's single-writer and DynamoDB's eventual consistency). Choose DSQL when: you need multi-region active-active writes WITH SQL and strong consistency — previously impossible on AWS without third-party (CockroachDB, Spanner).
3. What are S3 Tables? How do they change the traditional data lake architecture (S3 + Glue Catalog + Athena)?
   Answer Guide

   S3 Tables are built-in Apache Iceberg tables managed natively by S3. Eliminates Glue Crawler, Glue Catalog management, manual compaction, and metadata management. Up to 3x faster query performance and 10x more transactions/sec. Simplifies the data lakehouse stack from 4-5 services to S3 + Athena.
4. An interviewer asks: "What was the most impactful AWS launch from Re:Invent 2024 and why?" How would you answer this using the architectural significance framework?
   Answer Guide

   Structure your answer: "I'd pick [service] because it solves [problem] that previously required [workaround]. For example, Aurora DSQL solves multi-region strong consistency for SQL — previously you needed CockroachDB or accepted eventual consistency with DynamoDB. This changes how we architect global applications." Show you evaluate launches by architectural impact, not just hype.
5. Your existing EKS cluster uses Karpenter for node scaling with custom NodePools configured for GPU and Spot instances. Should you migrate to EKS Auto Mode? What factors influence this decision?
   Answer Guide

   Probably not — EKS Auto Mode doesn't support custom AMIs or GPU-specific node configurations (at launch). If you need: custom Karpenter NodePools, GPU workloads, specific instance type constraints, or Bottlerocket customization — stay with standard EKS + Karpenter. Auto Mode is best for standard compute workloads where simplicity outweighs control.
6. A data engineer asks whether they should migrate their existing Glue-based data pipeline to use S3 Tables. What questions would you ask before recommending?
   Answer Guide

   Ask: 1) Do you use Apache Iceberg already? (S3 Tables is Iceberg-native). 2) Do other tools query your data (Spark, Trino)? Iceberg format ensures compatibility. 3) How much operational overhead is Glue Crawler + Catalog? 4) What's the query performance requirement? 5) Is your data mostly structured tabular data? (S3 Tables is for tabular, not unstructured).
7. How do you stay current with AWS service launches? An interviewer wants to know your approach to continuous learning.
   Answer Guide

   Re:Invent keynotes and breakout sessions. AWS What's New feed (subscribe by category). AWS blogs (Architecture Blog, Database Blog). Follow AWS heroes and community builders. Hands-on: spin up new services in a sandbox account. The key: don't just read about services — evaluate them against your current architecture decisions.
8. Aurora DSQL offers strong consistency across regions. What is the performance trade-off? In which scenarios would the latency penalty make DSQL a poor choice?
   Answer Guide

   Strong consistency across regions requires cross-region coordination (consensus protocol), adding latency (potentially 100-300ms for cross-region writes vs single-digit ms for single-region Aurora). Poor choice for: latency-sensitive applications where all writes can be single-region, high-throughput write-heavy workloads, or when eventual consistency is acceptable (DynamoDB Global Tables would be faster and cheaper).