Skip to content

Errors by Type and Error Rate

When a run had failures, the Errors tab shows an Errors by type table above the chart: one row per (type, request) with the count and the share of all errors. The type is either an HTTP status code or, for failures that never got a response, the error the load generator recorded.

  • 4xx - the request was rejected: 401/403 authentication, 404 a wrong URL or a missing fixture, 429 rate limiting. These usually mean the script or the test data, not capacity.
  • 5xx - the server failed: 500 application errors, 502/503/504 an overloaded or restarting backend behind a proxy. These grow with load when capacity runs out.
  • SSLException / SSLHandshakeException - the TLS handshake failed, typically an expired certificate, a protocol mismatch or a server dropping handshakes under load.
  • HttpHostConnectException / ConnectException - no TCP connection at all: the host is down, a firewall drops the generator's IP, or the listen backlog is full.
  • SocketException / NoHttpResponseException / SocketTimeoutException - the connection opened but the server closed it or never answered: worker exhaustion, an upstream timeout, a keep-alive limit.

The table is already sorted by count, highest first. Under a type you also see the most frequent error message; a "+N" button lists the others. Open the request in the Stats table to see its response times.

Many hosting platforms put bot protection, DDoS protection or rate limits in front of a site, and these can refuse most of a load test. When a cloud test has finished and at least 20% of its requests ended with 403, 429, 430 (503 on Cloudflare) or a dropped or failed connection (SSLException, SSLHandshakeException, Connection reset, NoHttpResponseException), the results overview shows a notice such as "Vercel is probably filtering the test traffic." It names the platform when it recognises the host (for example Vercel, Netlify, Cloudflare, GitHub Pages, Firebase Hosting, Render, Heroku, Amazon CloudFront or Shopify), or says "This site appears to be rate limiting or blocking the test traffic." The notice changes no numbers. To measure the site itself, allow the test traffic in the platform's firewall or bot protection settings, using the IP ranges to allow, or test a staging deployment with that protection relaxed.

You can get an earlier hint while a cloud test is still starting. LoadFocus sends one test request to your site from AWS, and if a CDN or firewall it recognises (such as Cloudflare or Akamai) refuses it, the start-up screen shows a warning such as "Cloudflare turned away a test request to example.com from AWS eu-west-1. Your load test may be blocked the same way." It is only a warning: the test still runs, and a single request is not the load test, so read the results before you change anything.

The Error rate % toggle in the sidebar adds a line with the share of failed requests per second, computed from the individual samples. Read it together with the virtual-users line on the Timeline tab: an error rate that starts at a specific user count is a capacity limit; one that is constant from the first second is a script or environment problem.

See also common HTTP error codes.