Interview questions in Cloud Platforms, page 6
The small files problem in data lakes
Too many tiny files inflate metadata and per-file overhead, slowing queries; caused by streaming micro-batches and over-partitioning; fix with compaction and table formats like Iceberg, Delta, or Hudi.
Schema evolution without rewriting history
Use a table format with metadata-level evolution, add a new column rather than mutating the old, and reconcile types at read time; avoid rewriting petabytes.
Pre-built AI service vs custom model
Choose a managed service for speed, no ML expertise, and common tasks; build custom for domain-specific needs, control, or cost at scale.
Feeding large object-store data into training
Stream data instead of copying it all to disk, use streaming/pipe modes, shard and prefetch in parallel, and pack many small images into larger files.
Deploying a real-time inference endpoint
Package the model artifact and inference code in a container, choose instance type and autoscaling, configure the endpoint with health checks, and plan safe rollout like canary plus monitoring.
Inference performance bottlenecks on Lambda
Cold starts loading the model, memory and CPU limits, no GPU, and package size dominate; mitigate with provisioned concurrency, loading the model once outside the handler, smaller models, and right-sized…
How would you build CI/CD for an ML model?
Data and model versioning, automated training plus evaluation gates, model registry, deployment with monitoring and retraining triggers.
How would you speed up slow single-GPU training?
Vertical scaling to bigger or multi-GPU instances, then data-parallel or model-parallel distributed training across nodes.
Design auto drift detection and retraining
Capture inputs and predictions, compute data and concept drift metrics on a schedule, alert on threshold breach, and trigger a retraining and redeploy pipeline.
Design petabyte-scale distributed training
Object storage with columnar formats, distributed preprocessing, a data-parallel framework with efficient sharded loading, and managed orchestration.
Explain the Well-Architected Framework pillars
Name the pillars, operational excellence, security, reliability, performance efficiency, cost optimization, and sustainability, and explain each briefly.
Lift-and-shift versus replatforming migration
Rehosting moves apps unchanged for speed and low risk; replatforming makes targeted optimizations for cloud benefits; explain the speed-versus-value trade-off.
Design a highly available web application
Redundant stateless instances across multiple zones behind a load balancer with health checks, auto-scaling, and a replicated multi-AZ datastore.
Explain the Strangler Fig pattern
A facade routes traffic, new services gradually replace legacy features one slice at a time, and the old system is retired when fully strangled.
Managed services versus self-hosting trade-offs
Managed services cut operational burden and speed delivery but cost more and limit control; self-hosting offers full control and tuning at the price of patching, scaling, and reliability work.
Technical challenges of a multi-cloud strategy
Data consistency and egress costs across providers, cross-cloud networking and latency, and federating disparate IAM systems, plus operational and tooling overhead.
Design an enterprise cloud landing zone
Multi-account or subscription structure, centralized identity and SSO, network topology like hub-and-spoke, guardrails via policy and SCPs, and centralized logging.
Migrate an OLTP database with minimal downtime
Take an initial bulk load, then use change data capture to replicate ongoing changes until source and target are in sync, validate, then cut over during a brief window with a rollback plan.
Balance agility and compliance in regulated cloud
PaaS for speed where allowed, IaaS where control is required, enforced by encryption, IAM least privilege, network isolation, policy-as-code guardrails, and continuous audit logging.
When to choose bare metal over a VM
Bare metal suits latency-sensitive or high-throughput workloads needing no hypervisor overhead, single-tenant isolation for compliance, or direct hardware and licensing access.
We are hiring for this. Every open role lists the topics its interview covers, so you can prepare for the real thing rather than guessing.
See open roles