Skip to main content

Statistics

An instance of Statistics captures the runtime statistics of resolvers. The Statistics structure uses an online linear regression algorithm to compute the relationship between batch size and execution time, such that memory usage is minimal.

The Statistics object records a mapping from String to:

  • count: the number of points the regression contains.
  • meanX: the mean x coordinate of all the points.
  • meanY: the mean y coordinate of all the points.
  • varX: the variance of the x coordinates.
  • covXY: the covariance of the x and y coordinates.

The slope of the function can be computed as covXY / varX and the intercept as meanY - slope * meanX.

The intercept acts the cost of one element while the slope is the per element cost.

The intercept is the important of the two, since it allows us to compare batch resolvers regardless of their average batch sizes.

import cats.effect._
import gql._
import scala.concurrent.duration._
import cats.effect.unsafe.implicits.global

Statistics[IO].flatMap{ stats =>
  stats.updateStats("foo", 1.millis, 1) >>
    stats.updateStats("foo", 2.millis, 4) >>
    stats.updateStats("foo", 3.millis, 7) >>
    stats.updateStats("foo", 4.millis, 10) >>
    stats.getStats("foo")
}.unsafeRunSync()
// res0: Statistics.Stats = Stats(
//   initialCost = 1000.0,
//   extraElementCost = 333.3333333333333
// )