Troubleshooting CloudFront and Global Accelerator

A request that fails at the edge of a delivery network can fail in two very different ways, and the difference between them points to two entirely different layers. A fast failure, one that returns in milliseconds no matter how many times it's retried, usually means something at the edge is responding immediately, without ever reaching upstream. A slow failure, one that takes seconds and only shows up from certain locations, usually means the edge is waiting on something upstream and failing late. Those two symptoms look similar on a dashboard: error rate up, but they point at opposite ends of the request path.

Here's what that distinction looks like in practice. A customer reports two problems that sound similar but behave differently. One user gets a fast 404 that returns in tens of milliseconds no matter how many times they retry. Another user sees slow 504 errors that take several seconds and only happen from one geography.

Before touching any configuration, pinning down the exact path a request takes, and what can break at each boundary along that path, comes first.

That path is a chain, and each link leaves different evidence when it fails. A client that never reaches the edge shows up as the wrong hostname or wrong IPs in DNS answers, with no matching request records at ...