實驗?zāi)繕?/h5>

• 判斷NodePort內(nèi)部轉(zhuǎn)發(fā)是否負載均衡
• 判斷iptables和ipvs相同訪問的情況下能力差距
• 實驗環(huán)境除iptables和ipvs的模式以外，版本，配置，網(wǎng)絡(luò)插件，均一直

實驗步驟

1.創(chuàng)建多副本deployment分布到不通主機器（一定要保證pod分布均勻）
# kubectl create deployment nginx-test --image=nginx
# kubectl scale deployment nginx-test --replicas=9
# kubectl get deployment -o wide

2.創(chuàng)建SVC
# kubectl expose deployment nginx-test --type=NodePort --target-port=80 --port=80 --name=nginx-svc

3.修改相關(guān)index用于識別
# cat index.sh
for i in `kubectl get pod -o wide| grep -v NAME | awk '{print $1":"$7}'`;
do
  name=`echo $i | awk -F ":" '{print $1}'`
  host=`echo $i | awk -F ":" '{print $2}'`
  cmd=`echo 'kubectl exec -it $name -- bash -c "echo' $host '> /usr/share/nginx/html/index.html"'`
  eval $cmd
done

4.統(tǒng)計訪問100次都是那些主機
# cat sum.sh
log_file="/tmp/sum.txt"
echo > $log_file
start_time=`date "+%s"`
for i in {1..5000};
do
  curl -s http://192.168.26.104:32363/ >> $log_file
done
end_time=`date "+%s"`
echo "使用時間:" $[end_time-start_time]
cat /tmp/sum.txt | sort | uniq -c

python3版本測試用例

import requests,time

for_sum=5000
url="http://192.168.26.102:30030/"

sum_list={}
start_time=time.time()
for i in range(for_sum):
    resp=requests.get(url)
    try:
        sum_list[resp.content.decode()]=sum_list[resp.content.decode()] + 1
    except:
        sum_list[resp.content.decode()] = 0
end_time=time.time()
print("用時:", end_time - start_time  ,"秒")
print(sum_list)

環(huán)境記錄(svc值有一條,使用python3的測試代碼)

• iptables

用時: 23.67735481262207 秒
{'k8snode-1\n': 983, 'k8snode-2\n': 954, 'k8snode-5\n': 1014, 'k8snode-4\n': 1030, 'k8snode-3\n': 1014}

用時: 23.920344352722168 秒
{'k8snode-5\n': 1035, 'k8snode-1\n': 978, 'k8snode-3\n': 1002, 'k8snode-2\n': 1002, 'k8snode-4\n': 978}

用時: 23.591750144958496 秒
{'k8snode-5\n': 961, 'k8snode-1\n': 1017, 'k8snode-3\n': 1048, 'k8snode-2\n': 1011, 'k8snode-4\n': 958}

• ipvs

用時: 23.547069787979126 秒
{'k8snode-5\n': 999, 'k8snode-3\n': 999, 'k8snode-2\n': 999, 'k8snode-1\n': 999, 'k8snode-4\n': 999}

用時: 23.893048524856567 秒
{'k8snode-5\n': 999, 'k8snode-3\n': 999, 'k8snode-2\n': 999, 'k8snode-1\n': 999, 'k8snode-4\n': 999}

用時: 23.7876193523407 秒
{'k8snode-5\n': 999, 'k8snode-3\n': 999, 'k8snode-2\n': 999, 'k8snode-1\n': 999, 'k8snode-4\n': 999}

小結(jié)

使用iptables模式的nodeport存在負載不均衡，隨著iptables條數(shù)的增加導(dǎo)致不均衡的表現(xiàn)會體現(xiàn)的越加明顯