Each point compares a nightly build of the rapira main branch with the previous build on this board. The rig runs the two builds in pairs of cells, one after the other on the same instances. A point is the median change in percent over the pairs of the run: the capacity in req/s, and the p99 latency and the RSS at a constant request rate per target. The band shows the smallest and the largest change of the pairs. A higher capacity is better. A lower p99 and a lower RSS are better. A point is green for a better build and red for a worse build only when every round gave a pair, the whole band is on one side of zero, and the change is at least the noise floor of the measure. The gRPC p99 has a higher noise floor, because it changes more between two runs of the same build. All other points are gray. A gap is a run without a pair. The label of a run is the merged pull request of its commit. Hover over a chart, or tap it on a touch screen, to see the values of a run. A click opens the pull request, or the commit for a run without one. On a touch screen, a tap on a point opens it.