azure application gateway websockets latency metric
When using websockets together with Azure Application Gateway, you end up with artifically increased latency_d in the ApplicationGatewayPerformanceLog. Indeed, all the 101 (websockets) connections remain pending, which is a normal behavior and their duration gets recorded by the gateway. The problem is this normal behavior increases the average latency of all requests (including non-101) and there is no way to filter 101 out of the ApplicationGatewayPerformanceLog logs...Therefore, if we setup an alert on latency_d, this will raise a lot of false positives...While this metric is very useful in the ApplicationGatewayAccessLog because it allows for calculation of average user sessions, it is making the ApplicationGatewayPerformanceLog unreliable because there is no way we can discard 101 from this category.
Thanks for your feedback. We have added this item to our backlog.