Skip to content

Latest commit

 

History

History
136 lines (105 loc) · 6.01 KB

troubleshooting.md

File metadata and controls

136 lines (105 loc) · 6.01 KB
title description lead date draft images menu weight toc
Troubleshooting
Guide on troubleshooting the Prometheus Operator.
2021-03-08 08:49:31 +0000
false
docs
parent
operator
600
true

RBAC on Google Container Engine (GKE)

When you try to create ClusterRole (kube-state-metrics, prometheus prometheus-operator, etc.) on GKE Kubernetes cluster running 1.6 version, you will probably run into permission errors:

<....>
Error from server (Forbidden): error when creating 
"manifests/prometheus-operator/prometheus-operator-cluster-role.yaml": 
clusterroles.rbac.authorization.k8s.io "prometheus-operator" is forbidden: attempt to grant extra privileges:
<....>

This is due to the way Container Engine checks permissions. From Google Kubernetes Engine docs:

Because of the way Container Engine checks permissions when you create a Role or ClusterRole, you must first create a RoleBinding that grants you all of the permissions included in the role you want to create. An example workaround is to create a RoleBinding that gives your Google identity a cluster-admin role before attempting to create additional Role or ClusterRole permissions. This is a known issue in the Beta release of Role-Based Access Control in Kubernetes and Container Engine version 1.6.

To overcome this, you must grant your current Google identity cluster-admin Role:

# get current google identity
$ gcloud info | grep Account
Account: [myname@example.org]

# grant cluster-admin to your current identity
$ kubectl create clusterrolebinding myname-cluster-admin-binding --clusterrole=cluster-admin --user=myname@example.org
Clusterrolebinding "myname-cluster-admin-binding" created

Troubleshooting ServiceMonitor changes

When creating/deleting/modifying ServiceMonitor objects it is sometimes not as obvious what piece is not working properly. This section gives a step by step guide how to troubleshoot such actions on a ServiceMonitor object.

Overview of ServiceMonitor tagging and related elements

A common problem related to ServiceMonitor identification by Prometheus is related to an incorrect tagging, that does not match the Prometheus custom resource definition scope, or lack of permission for the Prometheus ServiceAccount to get, list, watch Services and Endpoints from the target application being monitored. As a general guideline consider the diagram below, giving an example of a Deployment and Service called my-app, being monitored by Prometheus based on a ServiceMonitor named my-service-monitor:

flow diagram

Note: The ServiceMonitor references a Service (not a Deployment, or a Pod), by labels and by the port name in the Service. This port name is optional in Kubernetes, but must be specified for the ServiceMonitor to work. It is not the same as the port name on the Pod or container, although it can be.

Has my ServiceMonitor been picked up by Prometheus?

ServiceMonitor objects and the namespace where they belong are selected by the serviceMonitorSelector and serviceMonitorNamespaceSelectorof a Prometheus object. The name of a ServiceMonitor is encoded in the Prometheus configuration, so you can simply grep whether it is present there. The configuration generated by the Prometheus Operator is stored in a Kubernetes Secret, named after the Prometheus object name prefixed with prometheus- and is located in the same namespace as the Prometheus object. For example for a Prometheus object called k8s one can find out if the ServiceMonitor named my-service-monitor has been picked up with:

kubectl -n monitoring get secret prometheus-k8s -ojson | jq -r '.data["prometheus.yaml.gz"]' | base64 -d | gunzip | grep "my-service-monitor"

Prometheus kubelet metrics server returned HTTP status 403 Forbidden

Prometheus is installed, all looks good, however the Targets are all showing as down. All permissions seem to be good, yet no joy. Prometheus pulling metrics from all namespaces expect kube-system, and Prometheus has access to all namespaces including kube-system.

Did you check the webhooks?

Issue has been resolved by amending the webhooks to use 0.0.0.0 instead of 127.0.0.1. Follow the below commands and it will update the webhooks which allows connections to all clusterIP's in all namespaces and not just 127.0.0.1.

Update the kubelet service to include webhook and restart:

KUBEADM_SYSTEMD_CONF=/etc/systemd/system/kubelet.service.d/10-kubeadm.conf
sed -e "/cadvisor-port=0/d" -i "$KUBEADM_SYSTEMD_CONF"
if ! grep -q "authentication-token-webhook=true" "$KUBEADM_SYSTEMD_CONF"; then
  sed -e "s/--authorization-mode=Webhook/--authentication-token-webhook=true --authorization-mode=Webhook/" -i "$KUBEADM_SYSTEMD_CONF"
fi
systemctl daemon-reload
systemctl restart kubelet

Modify the kube controller and kube scheduler to allow for reading data:

sed -e "s/- --address=127.0.0.1/- --address=0.0.0.0/" -i /etc/kubernetes/manifests/kube-controller-manager.yaml
sed -e "s/- --address=127.0.0.1/- --address=0.0.0.0/" -i /etc/kubernetes/manifests/kube-scheduler.yaml

Using textual port number instead of port name

The ServiceMonitor expects to use the port name as defined on the Service. So, using the Service example from the diagram above, we have this Service definition:

kind: Service
metadata:
  labels:
    k8s-app: my-app
  name: my-app
...
  spec:
    ports:
      - name: metrics
        port: 8080
    selector:
      k8s-app: my-app

We would then define the service monitor using metrics as the port not "8080". E.g.

CORRECT

kind: ServiceMonitor
metadata:
  name: my-app
spec:
...
  endpoints:
    - port: metrics

INCORRECT

kind: ServiceMonitor
metadata:
  name: my-app
spec:
...
  endpoints:
    - port: "8080"

The incorrect example will give an error along these lines spec.endpoints.port in body must be of type string: "integer"