What are the real pitfalls of running a Lambda-based serverless architecture at scale?
What they are really testing: Senior signal: do you know the operational sharp edges (cold starts, concurrency limits, downstream throttling, idempotency) rather than just the "no servers to manage" pitch.
A real interview question
What are the real pitfalls of running a Lambda-based serverless architecture at scale?
What most people say
drag me
“The main downside is cold starts adding some latency to the first request.”
Cold starts are real but the least of it at scale. The senior concerns, concurrency throttling, hammering a database, idempotency under at-least-once delivery, are missing entirely.
The follow-ups they ask next
Your Lambda functions are taking down your RDS database under load. Why, and what fixes it?
Lambda scales to many concurrent executions, each opening DB connections, exhausting RDS. Use RDS Proxy to pool/reuse connections, or buffer work through SQS so consumption is rate-limited.
Why must Lambda handlers be idempotent?
Async and stream/queue invocations are at-least-once, so the same event can be delivered more than once. Without a dedup key or idempotent logic, you double-process (double charges, duplicate records).
What the interviewer is listening for
- Goes beyond cold starts to concurrency/throttling
- Knows Lambda can overwhelm RDS (Proxy/SQS buffer)
- Insists on idempotency + DLQs
- Mentions Step Functions for orchestration
What sinks the answer
- Only cites cold starts
- Unaware of concurrency limits/throttling
- No idempotency under at-least-once delivery
If you genuinely do not know
Say this instead of freezing. Reasoning out loud from what you do know beats silence every single time, and a good interviewer is listening for exactly that.
“Cold starts [add startup latency, fix with provisioned concurrency + slim packages], but at scale the bigger ones: [concurrency limits cause throttling/429s, use reserved concurrency], [Lambda can exhaust RDS, fix with RDS Proxy or buffer via SQS], [at-least-once means handlers must be idempotent with DLQs], and [respect 15-min timeout/payload limits, orchestrate with Step Functions].”
Keep going with aws
Foundation
What are the EC2 purchasing options (On-Demand, Reserved, Savings Plans, Spot), and how do you mix them?
Foundation
Walk me through the main S3 storage classes and how you would use lifecycle policies to control cost.
Junior
How does AWS decide whether an IAM request is allowed? Explain the evaluation logic.
Junior
How does an EC2 Auto Scaling Group work, and how would you configure it to scale on load?
Mid
How do you choose between SQS, SNS, EventBridge, and Kinesis?
Mid
What are VPC endpoints, and when would you use a gateway endpoint versus an interface endpoint?
Knowing the answer is not the same as recalling it under pressure
Sign in to send the questions you fumble to spaced recall, so they come back right before you would forget them, and learn the concepts behind them with hands-on labs.
Start free