Initial scale kept for first minute of experiment

pkg/driver/deploy.sh

@@ -30,12 +30,13 @@ export FUNC_NAME=$2
 export CPU_REQUEST=$3
 export CPU_LIMITS=$4
 export MEMORY_REQUESTS=$5
-INIT_SCALE=$6
+export INIT_SCALE=$6
+export MIN_SCALE=$6 # minimum scale is the same as initial scale during profiling
 
 export PANIC_WINDOW=$7
 export PANIC_THRESHOLD=$8
 
 export AUTOSCALING_METRIC=$9
 export AUTOSCALING_TARGET=${10}
 
-cat $CONFIG_FILE | envsubst | kn service apply $FUNC_NAME --scale-init $INIT_SCALE --concurrency-target 1 --wait-timeout 2000000 -f /dev/stdin
+cat $CONFIG_FILE | envsubst | kn service apply $FUNC_NAME --concurrency-target 1 --wait-timeout 2000000 -f /dev/stdin

pkg/driver/deployment.go

@@ -55,6 +55,7 @@ func DeployFunctionsKnative(functions []*common.Function, yamlPath string, isPar
 
 func deployOne(function *common.Function, yamlPath string, isPartiallyPanic bool, endpointPort int,
 	autoscalingMetric string) bool {
+
 	panicWindow := "\"10.0\""
 	panicThreshold := "\"200.0\""
 	if isPartiallyPanic {
@@ -77,7 +78,7 @@ func deployOne(function *common.Function, yamlPath string, isPartiallyPanic bool
 		strconv.Itoa(function.CPURequestsMilli)+"m",
 		strconv.Itoa(function.CPULimitsMilli)+"m",
 		strconv.Itoa(function.MemoryRequestsMiB)+"Mi",
-		strconv.Itoa(function.InitialScale),
+		"\""+strconv.Itoa(function.InitialScale)+"\"",
 
 		panicWindow,
 		panicThreshold,
@@ -111,6 +112,39 @@ func deployOne(function *common.Function, yamlPath string, isPartiallyPanic bool
 	return true
 }
 
+func UpdateFunctionKnative(function *common.Function, endpointPort int) bool {
+
+	cmd := exec.Command(
+		"bash",
+		"./pkg/driver/update_deployment.sh",
+		function.Name,
+	)
+
+	stdoutStderr, err := cmd.CombinedOutput()
+	log.Debug("CMD response: ", string(stdoutStderr))
+
+	if err != nil {
+		// TODO: there should be a toggle to turn off deployment because if this is fatal then we cannot test the thing locally
+		log.Warnf("Failed to update deployment of function %s: %v\n%s\n", function.Name, err, stdoutStderr)
+
+		return false
+	}
+
+	if endpoint := urlRegex.FindStringSubmatch(string(stdoutStderr))[1]; endpoint != function.Endpoint {
+		// TODO: check when this situation happens
+		log.Debugf("Update function endpoint to %s\n", endpoint)
+		function.Endpoint = endpoint
+	} else {
+		function.Endpoint = fmt.Sprintf("%s.%s.%s", function.Name, namespace, bareMetalLbGateway)
+	}
+
+	// adding port to the endpoint
+	function.Endpoint = fmt.Sprintf("%s:%d", function.Endpoint, endpointPort)
+
+	log.Debugf("Updated deployment of function on %s\n", function.Endpoint)
+	return true
+}
+
 func CleanKnative() {
 	cmd := exec.Command("kn", "service", "delete", "--all")
 	var out bytes.Buffer

pkg/driver/trace_driver.go

@@ -255,10 +255,9 @@ func (d *Driver) individualFunctionDriver(function *common.Function, announceFun
 
 	if d.Configuration.WithWarmup() {
 		currentPhase = common.WarmupPhase
-		// skip the first minute because of profiling
-		minuteIndex = 1
+		minuteIndex = 0
 
-		log.Infof("Warmup phase has started.")
+		log.Info("Profiling phase for 1 minute, no downscaling below initial scale.")
 	}
 
 	startOfMinute := time.Now()
@@ -374,6 +373,12 @@ func (d *Driver) proceedToNextMinute(function *common.Function, minuteIndex *int
 	*invocationIndex = 0
 	*previousIATSum = 0
 
+	if d.Configuration.WithWarmup() && *minuteIndex == 1 {
+		log.Info("profiling phase is done, updating deployments to allow downscaling below initial scale.")
+		if d.Configuration.LoaderConfiguration.Platform == "Knative" {
+			UpdateFunctionKnative(function, d.Configuration.LoaderConfiguration.EndpointPort)
+		}
+	}
 	if d.Configuration.WithWarmup() && *minuteIndex == (d.Configuration.LoaderConfiguration.WarmupDuration+1) {
 		*currentPhase = common.ExecutionPhase
 		log.Infof("Warmup phase has finished. Starting the execution phase.")

pkg/driver/trace_driver_test.go

@@ -318,9 +318,9 @@ func TestDriverCompletely(t *testing.T) {
 				record := records[i]
 
 				if test.withWarmup {
-					if i < 5 && record.Phase != int(common.WarmupPhase) {
+					if i < 10 && record.Phase != int(common.WarmupPhase) {
 						t.Error("Invalid record phase in warmup.")
-					} else if i > 5 && record.Phase != int(common.ExecutionPhase) {
+					} else if i >= 10 && record.Phase != int(common.ExecutionPhase) {
 						t.Error("Invalid record phase in execution phase.")
 					}
 				}
@@ -342,7 +342,7 @@ func TestDriverCompletely(t *testing.T) {
 
 			expectedInvocations := 5
 			if test.withWarmup {
-				expectedInvocations = 10
+				expectedInvocations = 15
 			}
 
 			if !(successfulInvocation == expectedInvocations && failedInvocations == 0) {

pkg/driver/update_deployment.sh

@@ -0,0 +1,29 @@
+#!/usr/bin/env bash
+
+#
+# MIT License
+#
+# Copyright (c) 2023 EASL and the vHive community
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+#
+
+export FUNC_NAME=$1
+
+kn service update $FUNC_NAME --scale-min 0 --scale-init 0

pkg/trace/profiler.go

@@ -25,9 +25,10 @@
 package trace
 
 import (
+	"math"
+
 	log "github.com/sirupsen/logrus"
 	"github.com/vhive-serverless/loader/pkg/common"
-	"math"
 )
 
 func DoStaticTraceProfiling(functions []*common.Function) {
@@ -43,6 +44,7 @@ func ApplyResourceLimits(functions []*common.Function) {
 	for i := 0; i < len(functions); i++ {
 		memoryPct100 := int(functions[i].MemoryStats.Percentile100)
 		cpuShare := ConvertMemoryToCpu(memoryPct100)
+		// cpuShare := ConvertExecutionTimeToCpu(functions[i])
 
 		functions[i].CPURequestsMilli = cpuShare / common.OvercommitmentRatio
 		functions[i].MemoryRequestsMiB = memoryPct100 / common.OvercommitmentRatio
@@ -71,6 +73,26 @@ func ConvertMemoryToCpu(memoryRequest int) int {
 	return int(cpuRequest * 1000)
 }
 
+func ConvertExecutionTimeToCpu(function *common.Function) int {
+	var cpuRequest float32
+	switch executionTime := function.RuntimeStats.Average; {
+	case executionTime < 100:
+		cpuRequest = 1
+	case executionTime < 200:
+		cpuRequest = 0.9
+	case executionTime < 300:
+		cpuRequest = 0.8
+	case executionTime < 400:
+		cpuRequest = 0.7
+	case executionTime < 500:
+		cpuRequest = 0.6
+	default:
+		cpuRequest = 0.5
+	}
+
+	return int(cpuRequest * 1000)
+}
+
 func profileConcurrency(function *common.Function) float64 {
 	IPM := function.InvocationStats.Invocations[0]
 

workloads/container/trace_func_go.yaml

@@ -7,8 +7,8 @@ spec:
   template:
     metadata:
       annotations:
-        autoscaling.knative.dev/initial-scale: "0"  # Should start from 0, otherwise we can't deploy more functions than the node physically permits.
-        autoscaling.knative.dev/min-scale: "0"  # This parameter only has a per-revision key, so it's necessary to have here in case of the warmup messes up.
+        autoscaling.knative.dev/initial-scale: $INIT_SCALE # Should start from 0, otherwise we can't deploy more functions than the node physically permits.
+        autoscaling.knative.dev/min-scale: $MIN_SCALE  # This parameter only has a per-revision key, so it's necessary to have here in case of the warmup messes up.
         autoscaling.knative.dev/target-utilization-percentage: "100"  # Enforce container concurrency at any time.
         autoscaling.knative.dev/target-burst-capacity: "-1"  # Put activator always in the path explicitly.
         autoscaling.knative.dev/max-scale: "200"  # Maximum instances limit of Azure.