App service high disk queue length
We have observed high app service disk queue length while testing an application hosted on App service plan . As per investigation along with the microsoft support engineer, it has been identified that the disk queue length metric has always been aggregated as a sum (Total) for the samples captured from a worker. So any small change in this metric on the worker will cause this to spike rapidly. Also multiple workers will additionally cause this counter to grow even faster. In short, this is not an actual reflection of the actual disk queue length at any point in time since it is summing the values from the samples Many times per min in a typical case.
So, I would recommend to document this behavior which has always existed, or change the implementation to average the metric across the time granularity which we feel would be more useful as a customer.
We will add a work item to make these metrics clearer, though we don’t have a timeline we can share when this may get picked up.