Unable to access internet on pod in private GKE cluster

Question

I'm currently unable to access/ping/connect to any service outside of Google from my private Kubernetes cluster. The pods are running Alpine linux.

Routing Tables

/sleepez/api # ip route show table all
default via 10.52.1.1 dev eth0
10.52.1.0/24 dev eth0 scope link  src 10.52.1.4
broadcast 10.52.1.0 dev eth0 table local scope link  src 10.52.1.4
local 10.52.1.4 dev eth0 table local scope host  src 10.52.1.4
broadcast 10.52.1.255 dev eth0 table local scope link  src 10.52.1.4
broadcast 127.0.0.0 dev lo table local scope link  src 127.0.0.1
local 127.0.0.0/8 dev lo table local scope host  src 127.0.0.1
local 127.0.0.1 dev lo table local scope host  src 127.0.0.1
broadcast 127.255.255.255 dev lo table local scope link  src 127.0.0.1
local ::1 dev lo  metric 0
local fe80::ac29:afff:fea1:9357 dev lo  metric 0
fe80::/64 dev eth0  metric 256
ff00::/8 dev eth0  metric 256
unreachable default dev lo  metric -1  error -101

The pod certainly has an assigned IP and has no problem connecting to it's gateway:

PS C:\...\> kubectl get pods -o wide -n si-dev
NAME                              READY     STATUS    RESTARTS   AGE       IP          NODE
sleep-intel-api-79bf57bd9-c4l8d   1/1       Running   0          52m       10.52.1.4   gke-sez-production-default-pool-74b75ebc-6787

ip addr output

1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN qlen 1
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
3: eth0@if7: <BROADCAST,MULTICAST,UP,LOWER_UP,M-DOWN> mtu 1460 qdisc noqueue state UP
    link/ether 0a:58:0a:34:01:04 brd ff:ff:ff:ff:ff:ff
    inet 10.52.1.4/24 scope global eth0
       valid_lft forever preferred_lft forever
    inet6 fe80::ac29:afff:fea1:9357/64 scope link
       valid_lft forever preferred_lft forever

Pinging Gateway Works

/sleepez/api # ping 10.52.1.1
PING 10.52.1.1 (10.52.1.1): 56 data bytes
64 bytes from 10.52.1.1: seq=0 ttl=64 time=0.111 ms
64 bytes from 10.52.1.1: seq=1 ttl=64 time=0.148 ms
64 bytes from 10.52.1.1: seq=2 ttl=64 time=0.137 ms
^C
--- 10.52.1.1 ping statistics ---
3 packets transmitted, 3 packets received, 0% packet loss
round-trip min/avg/max = 0.111/0.132/0.148 ms

Pinging 1.1.1.1 Fails

/sleepez/api # ping 1.1.1.1
PING 1.1.1.1 (1.1.1.1): 56 data bytes
^C
--- 1.1.1.1 ping statistics ---
6 packets transmitted, 0 packets received, 100% packet loss

System Services Status

PS C:\...\> kubectl get deploy -n kube-system
NAME                    DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
event-exporter-v0.1.7   1         1         1            1           18m
heapster-v1.4.3         1         1         1            1           18m
kube-dns                2         2         2            2           18m
kube-dns-autoscaler     1         1         1            1           18m
l7-default-backend      1         1         1            1           18m
tiller-deploy           1         1         1            1           14m

Traceroute (Google Internal)

/sleepez/api # traceroute -In 74.125.69.105
 1  10.52.1.1  0.007 ms  0.006 ms  0.006 ms
 2  *  *  *
 3  *  *  *
 4  *  *

Traceroute (External)

traceroute to 1.1.1.1 (1.1.1.1), 30 hops max, 46 byte packets
 1  10.52.1.1  0.009 ms  0.003 ms  0.004 ms
 2  *  *  *
 3  *  *  *
 [continues...]

Solution

Nodes in a private GKE cluster do not have external IP addresses, so they cannot communicate with sites outside of Google. https://cloud.google.com/kubernetes-engine/docs/how-to/private-clusters#pulling_a_container_image_from_a_registry

Comments

[kasperd]

Including a traceroute from the VM to an IP address outside of Google as well as a traceroute from outside of Google to the external IP address of your VM will make this problem a lot easier to debug.

[Jenny]

@kasperd I included the traceroutes from inside the pod to an internal Google. I'm not sure how the external traceroute will help since it terminates at the Google-managed K8s cluster...

[kasperd]

The other traceroute I was asking for was from an external network to the external IP address of your VM.

[Jenny]

With a private GKE cluster the Compute nodes don't receive an external IP address. I can add one to provide the debugging info, but I still doubt that that is the issue

[Jenny]

This seems to be the correct answer and I wish this was highlighted a bit more in their documentation.

[Tarek]

u should use a nat gateway like Cloud NAT to redirect traffic in ur VPC through it. the cloud NAT runs outside ur cluster but on the same network.

[k_vishwanath]

@Tarek could you please provide the steps to enable NAT on private GKE.

[Jenny]

In retrospect to this question, I'm dumb, but I hope it helps others.