In order to determine the health of each backend, each Front Door environment periodically sends a synthetic HTTP/HTTPS request to each of your configured backends. Front Door then uses responses from these probes to determine the "best" backends to which it should route real client requests.
Front Door supports sending probes over either HTTP or HTTPS protocols. These probes are sent over the same TCP ports configured for routing client requests, and cannot be overridden.
Health probe responses
|Determining Health||A 200 OK status code indicates the backend is healthy. Everything else is considered a failure. If for any reason (including network failure) a valid HTTP response is not received for a probe, the probe is counted as a failure.|
|Measuring Latency||Latency is the wall-clock time measured from the moment immediately before we send the probe request to the moment when we receive the last byte of the response. We use a new TCP connection for each request, so this measurement is not biased towards backends with existing warm connections.|
How Front Door determines backend health
Azure Front Door Service uses the same three-step process below across all algorithms to determine health.
Exclude disabled backends.
Exclude backends that have health probes errors:
This selection is done by looking at the last n health probe responses. If at least x are healthy, the backend is considered healthy.
n is configured by changing the SampleSize property in load balancing settings.
x is configured by changing the SuccessfulSamplesRequired property in load balancing settings.
Out of the set of healthy backends in the backend pool, Front Door additionally measures and maintains the latency (round-trip time) for each backend.
Complete health probe failure
If health probes fail for every backend in a backend pool, then Front Door considers all backends healthy and routes traffic in a round robin distribution across all of them.
Once any backend returns to a healthy state, then Front Door will resume the normal load balancing algorithm.
Send feedback about: