API Gateway Latency Spikes Every 30 Minutes Like Clockwork

title: "API Gateway Latency Spikes Every 30 Minutes Like Clockwork" description: "If your latency spikes happen randomly, it's something else. If they happen every 25–30 minutes during low traffic, it's almost certainly Lambda cold starts." date: "2026-04-04" series: "aws-daily-with-divine" tags: ["aws", "lambda", "cold-starts", "serverless", "performance"] linkedinUrl: "https://www.linkedin.com/posts/divine-chukwu-63bb04145_aws-lambda-devops-activity-7443205245427417088-uAra"

Have you ever been a victim of AWS Lambda scaling to zero without knowing it was the cause of your latency issues?

Here's what's actually happening — and how to tell.

Lambda scales to zero

AWS Lambda automatically scales down to zero when there are no invocations. Functions that haven't been invoked for roughly 15 to 30 minutes get their execution environment shut down entirely.

The next invocation has to spin up a fresh environment, download the code, initialize the runtime, and run any initialization code in your function. That startup time is the cold start.

For simple Node or Python functions it might be 200–500ms. For Java or large packages, it can be several seconds.

How to diagnose it

This is the part most engineers miss while spending hours blaming the wrong thing.

• Latency spikes random or under heavy load? It's probably something else.
• Latency spikes like clockwork every 25–30 minutes during low traffic periods? It's almost certainly cold starts from your function scaling to zero.

The 30-minute pattern is the giveaway.

Two fixes, depending on your situation

Provisioned concurrency

Tell Lambda to keep a certain number of execution environments initialized and ready at all times so they never scale to zero. The first request always hits a warm environment with no cold start.

You pay for those environments even when they're idle, but you eliminate the latency spike completely.

aws lambda put-provisioned-concurrency-config \
  --function-name my-function \
  --qualifier prod \
  --provisioned-concurrent-executions 5

Scheduled warmer

Simpler alternative. A CloudWatch Events / EventBridge rule that pings your Lambda every 10 minutes keeps it warm without paying for provisioned concurrency.

aws events put-rule \
  --name lambda-warmer \
  --schedule-expression "rate(10 minutes)"
 
aws events put-targets \
  --rule lambda-warmer \
  --targets "Id"="1","Arn"="arn:aws:lambda:..."

Works well for low-traffic functions where you just need to prevent scale-to-zero without committing to reserved capacity.

Which to pick

Check your latency spikes before blaming anything else. Have you dealt with Lambda cold starts before? How did you solve it?