Question: Hey, any idea how to change "default backend - 404" on gce kubernetes? i did try
  - http:
    servicePort: 80
serviceName: test backend: got it working by adding
          servicePort: 80
but its still failing serviceName: test backend: - path: "/" paths:

Asked By
Asked At
2017-10-10 09:55:44

Found 15 possible answers.

User Answered At Possible Answer
alok 2017-10-10 11:53:35 @bboreham i was understanding how is pod to pod communication happening in the same node using weave. From what i know every pod has an ip and they talk to each other using a bridge in the same node. and they are created in their own network namesapces. I dont see any network namespace created in my node $ ip netns list gives me nothing. What am i doing wrong?
bboreham 2017-10-10 12:14:37 @alok ip netns list will only list namespaces that are linked into the filesystem in a particular way, and no container network follows that pattern hence you see nothing Each pod is given a virtual ethernet device (veth), where one end is in the pod's namespace and the other end is attached to the bridge in the host namespace. This is a common arrangement. For Weave Net specifically, pod-to-pod on the same node goes over a Linux bridge called weave For Kubernetes, every _pod_ is in its own network namespace If you do ls -l /proc/*/ns/net you will see every process' network namespace; they are symlinks to a magic name which gives a hint how they relate
alok 2017-10-10 13:28:50 thanks @bboreham
mcginnes 2017-10-10 13:44:50 Hi guys, we’re looking at doing some testing with ipvs on Kube 1.8. Our services are of Type NodePort, and I notice here it suggests iptables will be used for NodePort services - however when we started kube-proxy with --proxy-mode=ipvs --ipvs-min-sync-period 60s --ipvs-sync-period 300s --feature-gates=SupportIPVSProxyMode=true and created a service I can see with ipvsadm it has created some ipvs rules. Is this expected, and should we even be using ipvs if we have NodePort Services?
marc.zahn 2017-10-10 14:33:22 Hey guys - Just a question into the blue: I have a small Kubernetes cluster as playground. Network plugin is Calico. Now I have the problem that the pods from one host cannot interact with the cluster network - Neither dns nor communication with other pods. Communication with the "outer" world is possible. The cluster is set up with Kubespray. Does this sound familiar to someone?
lwr20 2017-10-10 14:35:34 @marc.zahn You may find more help here:
marc.zahn 2017-10-10 14:36:05 ok thx
song551 2017-10-10 14:49:50 @mcginnes FYI, I tested ipvs with NodePort services and found an issue about it.
mcginnes 2017-10-10 14:53:39 @song551 Looks similar to the issue we're having, although in our case I don't think it's getting to the pod at all
`` is our Service IP - but I don't know what `` is - it is not the IP address of the pod, which I'm guessing it should be, but I'm not sure..
-a -t -r -m -w 1
I'm not sure the ipvs rules are correct in our case - I use ipvsadm and can see it has generated rules such as:
-A -t -s rr
song551 2017-10-10 15:01:47 @mcginnes what network plugin you are using? Also what is your output of ipvsadm -L -n and ip a ?
mcginnes 2017-10-10 15:02:02 we're calico You want the full output from ip a?
  ->         Masq    1      0          3  
TCP rr
  ->         Masq    1      0          0         
TCP rr
song551 2017-10-10 15:10:58 No. That is fine. It is a bit strange from your ipvsadm output where the port number (30983) of the NodePort service has been shared by the backend pod.
mcginnes 2017-10-10 15:26:01 Would you expect should match the Endpoint IP address? As for port numbers this Endpoint does match, but we have 2 others in the same service which don't match, not sure if there's something special we do to get that match for that endpoint, but it always seems to match
andor 2017-10-10 15:32:08 does flannel suppor etcd3? cc @squeed never mind, should have googled first, tl;dr *no*
song551 2017-10-10 15:42:48 Yes. should match one of your backend pod id.

Related Questions