The Day DNS Broke Our Deployment: Solving a Serverless Framework S3 Resolution Failure on AWS

A developer encountered a complete deployment failure of a serverless AI application on AWS due to a DNS resolution error. The error EAI_AGAIN indicated a temporary DNS lookup failure on the local machine, not an AWS outage. Switching to reliable public DNS servers like 8.8.8.8 and 1.1.1.1 resolved the issue, and the team added safeguards such as retry logic and pre-deployment connectivity checks.

As engineers, we often spend hours optimizing code, improving prompts, and scaling infrastructure. But sometimes the biggest production issues come from something much simpler. A DNS lookup. Recently, while deploying a serverless AI application to AWS, I encountered an error that completely blocked deployment. The application hadn't changed. AWS was healthy. Permissions were correct. The deployment package was valid. Yet every deployment failed. The error looked like this: Error: getaddrinfo EAI AGAIN serverless-framework-deployments-eu-north-1-xxxxxxxx.s3.eu-north-1.amazonaws.com At first glance, it looked like an AWS outage. It wasn't. This is the story of how a simple DNS resolution issue brought an entire deployment pipeline to a halt—and how we fixed it. The application was an AI-powered Due Diligence Platform built with: Deployment flow: Developer ↓ Serverless Framework ↓ S3 Deployment Bucket ↓ CloudFormation ↓ Lambda Functions Every deployment package is first uploaded to an S3 bucket created by the Serverless Framework. Only after the upload succeeds does CloudFormation update the stack. During deployment, the terminal suddenly returned: serverless deploy ✖ Error: getaddrinfo EAI AGAIN serverless-framework-deployments-eu-north-1-xxxxxxxx.s3.eu-north-1.amazonaws.com The deployment stopped immediately. No Lambda updates. No CloudFormation changes. Nothing. The first thing I checked was AWS Service Health. Everything was operational. No incidents were reported. The next suspect was permissions. I verified: aws sts get-caller-identity Response: { "Account": "123456789012", "Arn": "arn:aws:iam::123456789012:user/developer" } Credentials were valid. Permissions were correct. Still failing. I upgraded Serverless Framework. npm install -g serverless Deployment still failed. The key clue was: EAI AGAIN This is not an AWS error. It is a DNS resolution error. Specifically: EAI AGAIN = Temporary DNS lookup failure The operating system could not resolve the S3 endpoint hostname. The request never reached AWS. I manually tested DNS resolution: nslookup google.com Intermittent failures appeared. Then: nslookup s3.eu-north-1.amazonaws.com The same issue occurred. This confirmed that the problem existed locally. Not in AWS. The machine was using an unstable DNS resolver. Under heavy network usage, DNS lookups occasionally timed out. When Serverless Framework attempted to upload artifacts to S3: Serverless ↓ DNS Lookup ↓ Failure ↓ Deployment Stops No connection to AWS was ever established. We switched to reliable public DNS servers. Linux: sudo nano /etc/resolv.conf Added: nameserver 8.8.8.8 nameserver 1.1.1.1 Then restarted networking: sudo systemctl restart NetworkManager After updating DNS: nslookup s3.eu-north-1.amazonaws.com Returned instantly. Deployment succeeded: serverless deploy Output: ✔ Service deployed successfully To avoid future issues, we added several safeguards. serverless deploy || serverless deploy Useful for CI/CD jobs when transient network issues occur. Before deployment: curl https://s3.eu-north-1.amazonaws.com If connectivity fails: Stop deployment This prevents wasting build minutes on doomed deployments. Added: aws sts get-caller-identity to deployment pipelines. This immediately detects expired or invalid credentials. The biggest lesson was simple: Not every AWS deployment error is actually an AWS problem. Sometimes: And the cloud gets blamed. When deployment issues occur, I now follow this order: aws sts get-caller-identity ping google.com nslookup s3.eu-north-1.amazonaws.com aws s3 ls serverless deploy This process has saved hours of troubleshooting. As engineers, we often expect complex problems to have complex causes. This incident reminded me that some of the most disruptive failures originate from the most basic layers of infrastructure. A single DNS lookup failure stopped an entire deployment pipeline. The code was correct. AWS was healthy. The architecture was sound. But none of that mattered until the network could resolve a hostname. Sometimes the fastest fix isn't changing code. It's understanding where the request actually fails. Before blaming AWS: You'll save yourself hours of debugging.