Revert shortfall scaling added in #40

scaler/scaling_calculator.go

@@ -33,25 +33,5 @@ func (sc *ScalingCalculator) DesiredCount(metrics *buildkite.AgentMetrics, asg *
 		desired = sc.perInstance(agentsRequired)
 	}
 
-	// If there are less agents registered than we'd expect based on the size
-	// of the autoscaling group, then there may be instances not accepting
-	// jobs: possibly because they are in the process of draining jobs for a
-	// graceful shutdown. In this case, we should expand the asg further to
-	// accommodate.
-	//
-	// Currently, restrict this behaviour to only occur when there are fewer
-	// than 2 instances (worth of agents) running. This will prevent any
-	// unexpected behaviour in large scaling groups. Since the effects of an
-	// instance not accepting jobs are more pronounced when there are fewer
-	// instances, this should cover the worst case scenario(s) where there may
-	// not be any instances accepting jobs.
-	if metrics.TotalAgents < int64(sc.agentsPerInstance) * 2 {
-		anticipated := (asg.DesiredCount - asg.Pending) * int64(sc.agentsPerInstance)
-		shortfall := anticipated - metrics.TotalAgents
-		if shortfall > 0 {
-			desired += sc.perInstance(shortfall)
-		}
-	}
-
 	return desired
 }