Autoscaler create pods less than requests.

[huasiy]

What version of Knative?

Kubernetes v1.29, Knative v1.13

0.9.x
0.10.x
0.11.x
Output of git describe --dirty

Expected Behavior

When I send five requests to existing knative service with seven running pods, five pods should be retained when scaling is triggered.

Actual Behavior

Only four pods will be retained. Sometimes I can even find four out seven pods deleted, one pod created immediately, thus reach four pods.

Steps to Reproduce the Problem

Run this script.

#! /bin/bash

set -ex

echo "Create the app"
cat > /tmp/service <<EOF
apiVersion: serving.knative.dev/v1
kind: Service
metadata:
  name: delete 
  namespace: default
spec:
  template:
    metadata:
      annotations:
        autoscaling.knative.dev/target: "1"
        autoscaling.knative.dev/target-utilization-percentage: "100"
        autoscaling.knative.dev/target-burst-capacity: "1"
        autoscaling.knative.dev/metric: "concurrency"
    spec:
      timeoutSeconds: 180 
      containers:
        - image: docker.io/hisy/delete:latest 
          ImagePullPolicy: IfNotPresent 
      terminationGracePeriodSeconds: 300
EOF

kn service apply -f /tmp/service 
sleep 5

export APP=$(kubectl get service.serving.knative.dev/delete | grep http | awk '{print $2}')

echo "Wait for pods to be terminated"
while [ $(kubectl get pods 2>/dev/null | wc -l) -ne 0 ];
do
  sleep 5;
done

echo "hit the autoscaler with burst of requests"
for i in `seq 7`; do
    curl -s "$APP?wait=10" 1>/dev/null &
done

echo "wait for the autoscaler to kick in and the bursty requests to finish"
sleep 30

echo "send longer requets"
for i in `seq 5`; do
    curl "$APP?wait=120"&
    sleep 1;
done

You can find seven pods created first. Then only four pods are retained.

[github-actions]

This issue is stale because it has been open for 90 days with no
activity. It will automatically close after 30 more days of
inactivity. Reopen the issue with /reopen. Mark the issue as
fresh by adding the comment /remove-lifecycle stale.