Troubleshooting
This document covers how to troubleshoot the deployment of a Kubernetes cluster, it will not cover debugging of workloads inside Kubernetes.
Understanding Cluster Status
Using juju status can give you some insight as to what’s happening in a cluster:
Model Controller Cloud/Region Version SLA Timestamp
charmed-kubernetes aws-91c aws/eu-west-1 2.4.5 unsupported 08:38:09+01:00
App Version Status Scale Charm Store Rev OS Notes
aws-integrator 1.15.71 active 1 aws-integrator jujucharms 5 ubuntu
easyrsa 3.0.1 active 1 easyrsa jujucharms 117 ubuntu
etcd 3.2.10 active 3 etcd jujucharms 209 ubuntu
flannel 0.10.0 active 5 flannel jujucharms 146 ubuntu
kubeapi-load-balancer 1.14.0 active 1 kubeapi-load-balancer jujucharms 162 ubuntu exposed
kubernetes-master 1.12.1 active 2 kubernetes-master jujucharms 219 ubuntu
kubernetes-worker 1.12.1 active 3 kubernetes-worker jujucharms 239 ubuntu exposed
Unit Workload Agent Machine Public address Ports Message
aws-integrator/0* active idle 0 54.171.121.229 ready
easyrsa/0* active idle 1 34.251.192.5 Certificate Authority connected.
etcd/0* active idle 2 52.18.186.65 2379/tcp Healthy with 3 known peers
etcd/1 active idle 3 54.194.35.197 2379/tcp Healthy with 3 known peers
etcd/2 active idle 4 34.240.14.183 2379/tcp Healthy with 3 known peers
kubeapi-load-balancer/0* active idle 5 34.244.110.15 443/tcp Loadbalancer ready.
kubernetes-master/0* active idle 6 34.254.175.71 6443/tcp Kubernetes master running.
flannel/0* active idle 34.254.175.71 Flannel subnet 10.1.16.1/24
kubernetes-master/1 active idle 7 52.210.61.51 6443/tcp Kubernetes master running.
flannel/3 active idle 52.210.61.51 Flannel subnet 10.1.38.1/24
kubernetes-worker/0* active idle 8 34.246.168.241 80/tcp,443/tcp Kubernetes worker running.
flannel/1 active idle 34.246.168.241 Flannel subnet 10.1.79.1/24
kubernetes-worker/1 active idle 9 54.229.236.169 80/tcp,443/tcp Kubernetes worker running.
flannel/4 active idle 54.229.236.169 Flannel subnet 10.1.10.1/24
kubernetes-worker/2 active idle 10 34.253.203.147 80/tcp,443/tcp Kubernetes worker running.
flannel/2 active idle 34.253.203.147 Flannel subnet 10.1.95.1/24
Entity Meter status Message
model amber user verification pending
Machine State DNS Inst id Series AZ Message
0 started 54.171.121.229 i-0f47fcfb452fa8fab bionic eu-west-1a running
1 started 34.251.192.5 i-011007983db6d2736 bionic eu-west-1b running
2 started 52.18.186.65 i-0b411be2a3909ae32 bionic eu-west-1a running
3 started 54.194.35.197 i-0fccba854c6d59ffe bionic eu-west-1b running
4 started 34.240.14.183 i-02148162336e08864 bionic eu-west-1c running
5 started 34.244.110.15 i-08833b743ebcd0d9c bionic eu-west-1c running
6 started 34.254.175.71 i-0f18d3f7377ba406f bionic eu-west-1a running
7 started 52.210.61.51 i-08ec1daf25fb18fa3 bionic eu-west-1b running
8 started 34.246.168.241 i-0934f74bfdfba2a3f bionic eu-west-1b running
9 started 54.229.236.169 i-0a4129c834c713a5e bionic eu-west-1a running
10 started 34.253.203.147 i-053492139b1080ce0 bionic eu-west-1c running
In this example we can glean some information. The Workload column will show the status of a given service. The Message section will show you the health of a given service in the cluster. During deployment and maintenance these workload statuses will update to reflect what a given node is doing. For example the workload may say maintenance while message will describe this maintenance as Installing docker.
During normal operation the Workload should read active, the Agent column (which reflects what the Juju agent is doing) should read idle, and the messages will either say Ready or another descriptive term. juju status --color will also return all green results when a cluster’s deployment is healthy.
Status can become unwieldy for large clusters, it is then recommended to check status on individual services, for example to check the status on the workers only:
juju status kubernetes-worker
or just on the etcd cluster:
juju status etcd
Errors will have an obvious message, and will return a red result when used with juju status --color. Nodes that come up in this manner should be investigated.
SSHing to units
You can ssh to individual units easily with the following convention, juju ssh <servicename>/<unit#>:
juju ssh kubernetes-worker/3
Will automatically ssh you to the 3rd worker unit.
juju ssh easyrsa/0
This will automatically ssh you to the easyrsa unit.
Collecting debug information
To collect comprehensive debug output from your Charmed Kubernetes cluster, install and run juju-crashdump on a computer that has the Juju client installed, with the current controller and model pointing at your Charmed Kubernetes deployment.
sudo snap install juju-crashdump --classic --channel edge
juju-crashdump -a debug-layer -a config
Running the juju-crashdump script will generate a tarball of debug information that includes systemd unit status and logs, Juju logs, charm unit data, and Kubernetes cluster information. It is recommended that you include this tarball when filing a bug.
Common Problems
Charms deployed to LXD containers fail after upgrade/reboot
For deployments using Juju’s localhost cloud, which deploys charms to LXD/LXC containers, or other
cases where applications are deployed to LXD, there is a known issue
(https://bugs.launchpad.net/juju/+bug/1904619)
with the profiles applied by Juju. The LXD profile used by Juju is named after the charm, including
the revision number. Upgrading the charm causes Juju to create a new profile for LXD which does not
necessarily contain the same settings which were originally supplied. If services based on LXD
containers fail to resume after an upgrade, this is a potential cause of that failure.
To check what the profiles should contain, the YAML output from juju status or juju machines can be used:
juju machines --format=yaml
… will detail the profiles in the output, e.g.:
model:
name: default
machines:
"0":
...
instance-id: juju-4ac678-1
...
lxd-profiles:
juju-default-kubernetes-worker-718:
config:
linux.kernel_modules: ip_tables,ip6_tables,netlink_diag,nf_nat,overlay
raw.lxc: |
lxc.apparmor.profile=unconfined
lxc.mount.auto=proc:rw sys:rw
lxc.cgroup.devices.allow=a
lxc.cap.drop=
security.nesting: "true"
security.privileged: "true"
description: ""
devices:
aadisable:
path: /dev/kmsg
source: /dev/kmsg
type: unix-char
...
To check this matches with the actually applied profile, you can run lxc (this needs to be run on the machine where the containers are running).
lxc profile show juju-default-kubernetes-worker-718
This should give the appropriate corresponding output:
config:
linux.kernel_modules: ip_tables,ip6_tables,netlink_diag,nf_nat,overlay
raw.lxc: |
lxc.apparmor.profile=unconfined
lxc.mount.auto=proc:rw sys:rw
lxc.cgroup.devices.allow=a
lxc.cap.drop=
security.nesting: "true"
security.privileged: "true"
description: ""
devices:
aadisable:
path: /dev/kmsg
source: /dev/kmsg
type: unix-char
name: juju-default-kubernetes-worker-718
used_by:
- /1.0/instances/juju-4ac678-1
If this differs from what is expected, the profile can be manually edited. E.g., for the above profile:
lxc profile edit juju-default-kubernetes-worker-718
Load Balancer interfering with Helm
This section assumes you have a working deployment of Kubernetes via Juju using a Load Balancer for the API, and that you are using Helm to deploy charts.
To deploy Helm you will have run:
helm init
$HELM_HOME has been configured at /home/ubuntu/.helm
Tiller (the helm server side component) has been installed into your Kubernetes Cluster.
Happy Helming!
Then when using helm you may see one of the following errors:
- Helm doesn’t get the version from the Tiller server
helm version
Client: &version.Version{SemVer:"v2.1.3", GitCommit:"5cbc48fb305ca4bf68c26eb8d2a7eb363227e973", GitTreeState:"clean"}
Error: cannot connect to Tiller
- Helm cannot install your chart
helm install <chart> --debug
Error: forwarding ports: error upgrading connection: Upgrade request required
This is caused by the API load balancer not forwarding ports in the context of the helm client-server relationship. To deploy using helm, you will need to follow these steps:
-
Expose the Kubernetes Master service
juju expose kubernetes-master -
Identify the public IP address of one of your masters
juju status kubernetes-masterModel Controller Cloud/Region Version SLA Timestamp charmed-kubernetes aws-91c aws/eu-west-1 2.4.5 unsupported 08:39:23+01:00 App Version Status Scale Charm Store Rev OS Notes flannel 0.10.0 active 2 flannel jujucharms 146 ubuntu kubernetes-master 1.12.1 active 2 kubernetes-master jujucharms 219 ubuntu Unit Workload Agent Machine Public address Ports Message kubernetes-master/0* active idle 6 34.254.175.71 6443/tcp Kubernetes master running. flannel/0* active idle 34.254.175.71 Flannel subnet 10.1.16.1/24 kubernetes-master/1 active idle 7 52.210.61.51 6443/tcp Kubernetes master running. flannel/3 active idle 52.210.61.51 Flannel subnet 10.1.38.1/24 Entity Meter status Message model amber user verification pending Machine State DNS Inst id Series AZ Message 6 started 34.254.175.71 i-0f18d3f7377ba406f bionic eu-west-1a running 7 started 52.210.61.51 i-08ec1daf25fb18fa3 bionic eu-west-1b runningIn this context the public IP address is 54.210.100.102.
If you want to access this data programmatically you can use the JSON output:
juju show-status kubernetes-master --format json | jq --raw-output '.applications."kubernetes-master".units | keys[]' 54.210.100.102 -
Update the kubeconfig file
Identify the kubeconfig file or section used for this cluster, and edit the server configuration.
By default, it will look like
https://54.213.123.123:443. Replace it with the Kubernetes Master endpointhttps://54.210.100.102:6443and save.Note that the default port used by Charmed Kubernetes for the Kubernetes Master API is 6443 while the port exposed by the load balancer is 443.
-
Start helm again!
helm install <chart> --debug Created tunnel using local port: '36749' SERVER: "localhost:36749" CHART PATH: /home/ubuntu/.helm/<chart> NAME: <chart> ... ...
Logging and monitoring
By default there is no log aggregation of the Kubernetes nodes, each node logs locally. Please read over the logging page for more information.
Troubleshooting Keystone/LDAP issues
The following section offers some notes to help determine issues with using Keystone for authentication/authorisation.
Testing the steps is important to determine the cause of the problem.
Can you communicate with Keystone and get an authorization token?
First is to verify that Keystone communication works from both your client and
the kubernetes-worker machines. The easiest thing to do here is to copy the
kube-keystone.sh script to the machines of interest from kubernetes-master with
juju scp kubernetes-master/0:kube-keystone.sh ., edit the script to include
your credentials, source kube-keystone.sh and then run get_keystone_token.
This will produce a token from the Keystone server. If that isn’t working,
check firewall settings on your Keystone server. Note that the
kube-keystone.sh script could be overwritten, so it is a best practice to make
a copy somewhere and use that.
Are the pods for Keystone authentication up and running properly?
The Keystone pods live in the kube-system namespace and read a configmap from Kubernetes for the policy. Check to make sure they are running:
kubectl -n kube-system get po
NAME READY STATUS RESTARTS AGE
k8s-keystone-auth-5c6b7f9b7c-mvvkx 1/1 Running 0 21m
k8s-keystone-auth-5c6b7f9b7c-q2jfq 1/1 Running 0 21m
Check the logs of the pods for errors:
kubectl -n kube-system logs k8s-keystone-auth-5c6b7f9b7c-mvvkx
W1121 05:02:02.878988 1 config.go:73] Argument --sync-config-file or --sync-configmap-name missing. Data synchronization between Keystone and Kubernetes is disabled.
I1121 05:02:02.879139 1 keystone.go:527] Creating kubernetes API client.
W1121 05:02:02.879151 1 client_config.go:548] Neither --kubeconfig nor --master was specified. Using the inClusterConfig. This might not work.
I1121 05:02:02.893499 1 keystone.go:544] Kubernetes API client created, server version v1.12
I1121 05:02:02.998944 1 keystone.go:93] ConfigMaps synced and ready
I1121 05:02:02.999045 1 keystone.go:101] Starting webhook server...
I1121 05:02:02.999262 1 keystone.go:155] ConfigMap created or updated, will update the authorization policy.
I1121 05:02:02.999459 1 keystone.go:171] Authorization policy updated.
Is the configmap with the policy correct?
Check the configmap contents. The pod logs above would complain if the YAML isn’t valid, but make sure it matches what you expect.
kubectl -n kube-system get configmap k8s-auth-policy -o=yaml
apiVersion: v1
data:
policies: |
[
{
"resource": {
"verbs": ["get", "list", "watch"],
"resources": ["pods"],
"version": "*",
"namespace": "default"
},
"match": [
{
"type": "user",
"values": ["admin"]
},
]
}
]
kind: ConfigMap
metadata:
annotations:
kubectl.kubernetes.io/last-applied-configuration: |
{"apiVersion":"v1","data":{"policies":"[\n {\n \"resource\": {\n \"verbs\": [\"get\", \"list\", \"watch\"],\n \"resources\": [\"pods\"],\n \"version\": \"*\",\n \"namespace\": \"default\"\n },\n \"match\": [\n {\n \"type\": \"user\",\n \"values\": [\"admin\"]\n },\n ]\n }\n]\n"},"kind":"ConfigMap","metadata":{"annotations":{},"labels":{"k8s-app":"k8s-keystone-auth"},"name":"k8s-auth-policy","namespace":"kube-system"}}
creationTimestamp: 2018-11-21T02:38:12Z
labels:
k8s-app: k8s-keystone-auth
name: k8s-auth-policy
namespace: kube-system
resourceVersion: "16736"
selfLink: /api/v1/namespaces/kube-system/configmaps/k8s-auth-policy
uid: 7dc0842b-ed36-11e8-82e1-06d4a9ac9e06
Check the service and endpoints
Verify the service exists and has endpoints
kubectl get svc -n kube-system
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
heapster ClusterIP 10.152.183.49 <none> 80/TCP 136m
**k8s-keystone-auth-service ClusterIP 10.152.183.200 <none> 8443/TCP 105m**
kube-dns ClusterIP 10.152.183.218 <none> 53/UDP,53/TCP 137m
kubernetes-dashboard ClusterIP 10.152.183.142 <none> 443/TCP 136m
metrics-server ClusterIP 10.152.183.245 <none> 443/TCP 136m
monitoring-grafana ClusterIP 10.152.183.2 <none> 80/TCP 136m
monitoring-influxdb ClusterIP 10.152.183.172 <none> 8083/TCP,8086/TCP 136m
$ kubectl -n kube-system get ep
NAME ENDPOINTS AGE
heapster 10.1.20.4:8082 136m
**k8s-keystone-auth-service 10.1.20.5:8443,10.1.32.4:8443 105m**
kube-controller-manager <none> 136m
kube-dns 10.1.31.6:53,10.1.31.6:53 136m
kube-scheduler <none> 136m
kubernetes-dashboard 10.1.31.3:8443 136m
metrics-server 10.1.32.2:443 136m
monitoring-grafana 10.1.31.2:3000 136m
monitoring-influxdb 10.1.31.2:8086,10.1.31.2:8083 136m
Attempt to authenticate directly to the service
Use a token to auth with the Keystone service directly:
cat <<EOF | curl -ks -XPOST -d @- https://10.152.183.200:8443/webhook | python -mjson.tool
{
"apiVersion": "authentication.k8s.io/v1beta1",
"kind": "TokenReview",
"metadata": {
"creationTimestamp": null
},
"spec": {
"token": "$(get_keystone_token)"
}
}
EOF
{
"apiVersion": "authentication.k8s.io/v1beta1",
"kind": "TokenReview",
"metadata": {
"creationTimestamp": null
},
"spec": {
"token": "gAAAAABb9Yeel_62KoSb_fAL6RPMpGZ4-4y5RLqXq5YdY3PcIKpuIcZ8PoVPhQtHOR7fiPYpFQX_pAUZJ4yngSE_WbJeuX8c-pl5WgStNImmkH3sEvQ5nSfimGhQSH-k5ydCBhcor87AeN7dOS-X6zHMRrcyvnZffQ"
},
"status": {
"authenticated": true,
"user": {
"extra": {
"alpha.kubernetes.io/identity/project/id": [
""
],
"alpha.kubernetes.io/identity/project/name": [
""
],
"alpha.kubernetes.io/identity/roles": [],
"alpha.kubernetes.io/identity/user/domain/id": [
"e1cbddf1b75340499109f0b88b28d472"
],
"alpha.kubernetes.io/identity/user/domain/name": [
"admin_domain"
]
},
"groups": [
""
],
"uid": "432f311e7eb94689b10aee03293ab030",
"username": "admin"
}
}
}
Note that you need to change the IP address above to the address of your
k8s-keystone-auth-service. This will talk to the webhook and verify that the token is
valid and return information about the user.
API server
Finally, communication between the API server and the Keystone service is verified. The easiest thing to do here is to look at the log for the API server for interesting information such as timeouts or errors with the webhook.
juju run --unit kubernetes-master/0 -- journalctl -u snap.kube-apiserver.daemon.service
We appreciate your feedback on the documentation. You can edit this page or file a bug here.