I have a k8s cluster, and the version of k8s is v1.12.1. One of the services is below:
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
service/flask NodePort 10.99.101.226 <none> 8090:32010/TCP 15h
When accessing the service with CLUSTER-IP inside the cluster, it is very slow compared to accessing inside the pod container.
curl 10.99.101.226:8090/test_json > /dev/null
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 2316k 100 2316k 0 0 263k 0 0:00:08 0:00:08 --:--:-- 216kkubectl exec -it flask-example-77dbd59654-2g857 -- sh
/app # curl localhost:8090/test_json > /dev/null
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 2316k 100 2316k 0 0 75.3M 0 --:--:-- --:--:-- --:--:-- 75.3MThe former speed is only 216K, but the latter speed can up to 75.3M, I wonder to know what cause the difference.
Is there anyone can help me? thanks.
There's the latency of your network connection.
When you access it by exec'ing into the pod, there's no hop at all as you are on the same host and network, but the NodePort service expose your app through your node, so your connection will need to go first from your machine to the Node and then bypassed to the pod.
[your machine] <==> [NODE]:32010 <==> [POD]:8090
To have a clear view on what's happening, you can use --trace and you should be able to see the hops:
curl --trace trace.log 10.99.101.226:32010/test_json > /dev/nullthen check the file trace.log
To avoid hit the node and access the pod directly, you can use a headless service. You can do that by changing the service type to ClusterIP and set clusterIP: None on the spec