We changed how we aggregate response time metrics in Judoscale. Previously we calculated an average response time every 10 seconds, and we’re now calculating a median response time every 10 seconds.

We’ve been exploring this data for several weeks, and we’ve found the median to be a better representation of overall capacity. It’s less prone to being skewed by one or two slow web endpoints which made the average response time more “spiky”.

Note: this does not affect apps using queue time metric, reported from Judoscale adapters.