[Clusterlint](https://github.com/digitalocean/clusterlint) flags issues with workloads deployed in a cluster. These issues might cause workload downtime during maintenance or upgrade to a new Kubernetes version, and could complicate the maintenance or upgrade itself.
Namespaces are a way to limit the scope of the resources that subsets of users within a team can create. While a default namespace is created for every Kubernetes cluster, we don't recommend adding all created resources into the default namespace because of the risk of privilege escalation, resource name collisions, latency in operations as resources scale up, and mismanagement of Kubernetes objects. Having namespaces lets you enable resource quotas can be enabled to track node, CPU and memory usage for individual teams.
We don't recommend using container images with the `latest` tag or not specifying a tag in the image (which defaults to `latest`), as this leads to confusion around the version of image used. Pods get rescheduled often as conditions inside a cluster change, and upon a reschedule, you may find that the images' versions have changed to use the latest release, which can break the application and make it difficult to debug errors. Instead, update segments of the application individually using images pinned to specific versions.
We do not recommend having a `concurrencyPolicy` of `Allow` for CronJob resources. If a CronJob-managed Pod does not execute to completion within the expected window, it is possible that multiple Pods pile up over time, leading to several Pods stuck in a pending state and possible resource contention. Instead, prefer `Forbid`, which skips execution of a new job if the previous job has not exited, or `Replace`, which replaces the still-running job with a new job if it has not yet exited.
### Example
```yaml
# Not recommended: Having a concurrency policy of Allow
apiVersion: batch/v1beta1
kind: CronJob
metadata:
name: mycron
spec:
concurrencyPolicy: Allow
```
### How to Fix
```yaml
# Recommended: Having a concurrency policy of Forbid or Replace
Use the `privileged` mode for trusted containers only. Because the privileged mode allows container processes to access the host, malicious containers can extensively damage the host and bring down services on the cluster. If you need to run containers in privileged mode, test the container before using it in production. For more information about the risks of running containers in privileged mode, please refer to the [Kubernetes security context documentation](https://kubernetes.io/docs/tasks/configure-pod-container/security-context/).
If containers within a pod are allowed to run with the process ID (PID) `0`, then the host can be subjected to malicious activity. We recommend using a user identifier (UID) other than `0` in your container image for running applications. You can also enforce this in the Kubernetes pod configuration as shown below.
Docker is the most popular runtime for Kubernetes. However, Kubernetes supports other container runtimes as well, such as containerd and CRI-O. If the registry is not prepended to the image name, docker assumes `docker.io` and pulls it from Docker Hub. However, the other runtimes will result in errors while pulling images. To maintain portability, we recommend using a fully qualified image name. If the underlying runtime is changed and the object configs are deployed to a new cluster, having fully qualified image names ensures that the applications don't break.
On upgrade of a cluster on DOKS, the worker nodes' hostname changes. So, if a user's pod spec relies on the hostname to schedule pods on specific nodes, pod scheduling will fail after the upgrade.
Admission control webhooks can disrupt upgrade and node replacement operations by preventing system components from starting. Specifically, this happens when an admission control webhook:
* has failurePolicy set to Fail,
* targets a service other than the Kubernetes apiserver, and
* applies to both kube-system and the namespace of the targeted service.
Admission control webhook timeouts can block upgrades, when the API call times out, due to an incorrectly configured TimeoutSeconds value. Since webhooks inherently add to API latency, we must stay within the recommended range in order for API requests to be successful. Specifically, this happens when an admission control webhook does not respond within 30 seconds.
### Example
```yaml
# Error: Configure a webhook with a TimeoutSeconds value greater than 30 seconds.
apiVersion: admissionregistration.k8s.io/v1beta1
kind: ValidatingWebhookConfiguration
metadata:
name: sample-webhook.example.com
webhooks:
- name: sample-webhook.example.com
rules:
- apiGroups:
- ""
apiVersions:
- v1
operations:
- CREATE
resources:
- pods
scope: "Namespaced"
clientConfig:
service:
namespace: webhook
name: webhook-server
path: /pods
admissionReviewVersions:
- v1beta1
timeoutSeconds: 60
```
### How to Fix
Set the TimeoutSeconds value to anything within the 1 to 30 second range.
DOBS pod owner check ensures that any pod that references a DO Block Storage volume is owned by a StatefulSet. We want such pods to be owned by a StatefulSet because:
1. The Eviction API does not respect deployment strategies. It only cares about pod disruption budgets (PDBs). So, if you don’t set it right, you can end up with multiple DOBS-using pods running concurrently.
This can lead to stuck deployments if they happen to come up on different nodes in the best case, and data corruption if they come up on the same node and end up writing to same volume concurrently. For more context, see: https://kubernetes.io/docs/concepts/workloads/pods/disruptions/.
2. Manual deletes do not care about PDBs at all. So, all pods from a Deployment, for instance are deleted and brought up at the same time. A StatefulSet, on the other hand, always ensures “at most” guarantees.
### Example
### Example
```yaml
# Not recommended: Pods that refer to DOBS volumes should be owned by a StatefulSet
apiVersion: v1
kind: Pod
metadata:
name: mypod
namespace: test
labels:
name: mypod
spec:
containers:
- name: mypod
image: nginx:1.17.0
volumeMounts:
- mountPath: "/data"
name: all-data
volumes:
- name: my-dobs-volume
persistentVolumeClaim:
claim-name: do-csi-pvc
```
### How to fix
```yaml
# Recommended: Ensure that a StatefulSet is used to configure pods referencing DOBS volumes
- Pods with an identical configuration (such as those created from a `podTemplate`) intended to behave identically to one another regardless of their deployment will in fact behave differently from node to node due to differences in the files present on the nodes themselves.
- Resource-aware scheduling is not be able to account for resources used by a `hostPath` volume.
- The files created on the hosts are only writable by root; you will need to run your process as root in a privileged container or modify the file permissions on the host to be able to write to a `hostPath` volume.
For more details about `hostPath` volumes, please refer to [the Kubernetes documentation](https://kubernetes.io/docs/concepts/storage/volumes/#hostpath)
This check reports all the persistent volumes in the cluster that are not claimed by a `PersistentVolumeClaim` (PVC) in any namespace. You can clean up the cluster based on this information and there will be fewer objects to manage.
This check reports all the PVCs in the cluster that are not referenced by pods in the respective namespaces. You can clean up the cluster based on this information.
This check reports all the config maps in the cluster that are not referenced by pods in the respective namespaces. You can clean up the cluster based on this information.
This check reports all the secret names in the cluster that are not referenced by pods in the respective namespaces. You can clean up the cluster based on this information.
When you specify resource limits for containers, the scheduler can make better decisions about which nodes to place pods on, and handle contention for resources on a node in a specified manner.
When the node that a pod is running on reboots or fails, the pod is terminated and will not be restarted. However, a job will create new pods to replace terminated ones. For this reason, we recommend that you use a job, deployment, or `StatefulSet` rather than a bare pod, even if your application requires only a single pod.
When a DOKS cluster is upgraded, all worker nodes are replaced, and replacement nodes do not retain any custom labels or taints that were previously set by the user on the nodes. This check reports any labels or taints that will be lost on upgrade.
DOKS provides persistent node pool labels. Adding a custom label to a node pool will ensure that the label is propagated to the worker nodes in the node pool after replacement or upgrade.