< img height="1" width="1" style="display:none;" alt="" src="https://px.ads.linkedin.com/collect/?pid=3131724&fmt=gif" />
Forest-L, Sherlock
Published on๏ผš2021-02-03    The number of views:

KubeEye: An Automatic Diagnostic Tool that Provides a Holistic View of Your Kubernetes Cluster

KubeEye is an open-source diagnostic tool for identifying various Kubernetes cluster issues automatically, such as misconfigurations, unhealthy components and node failures. It empowers cluster operators to manage and troubleshoot clusters in a timely and graceful manner. Developed in Go on the basis of Polaris and Node Problem Detector, KubeEye is equipped with a series of built-in rules for exception detection. Besides pre-defined rules, KubeEye also supports customized rules.


Why Open Source

As the de facto standard in container orchestration, Kubernetes features a powerful and flexible architecture which supports various cloud-native add-ons. That said, there are some obstacles and problems which have haunted cluster operators and engineers for so long, such as:

  • CPU, memory or disk exceptions

  • Kernel deadlock and filesystem damage

  • Container runtime daemon not responding

  • Unhealthy etcd

  • Request and limit blackbox

  • Pod failures

  • Certificates expiring

  • Docker service failures

  • OOM (Out of Memory)

In fact, the problems listed here are just the tip of the iceberg. The majority of the cluster issues lies below the surface as they are not directly detectable from the cluster control plane. In other words, even with these potential issues, Kubernetes may continue to schedule Pods onto malfunctioning nodes, thus posing a greater threat to the stability and security of your cluster.

To identify these issues in time, the KubeSphere team developed KubeEye with over three years of expertise in Kubernetes cluster management and maintenance in production for both community users and commercial customers.

What Can KubeEye Do

  • Identify and detect problems in your Kubernetes cluster control plane, including kube-apiserver, kube-controller-manager and etcd.

  • Identify and detect various node problems, including memory, CPU and disk pressure, and unexpected kernel issues.

  • Validate YAML specifications of your workloads for best practices and compliance in the industry, thus improving cluster stability.

KubeEye Architecture

By calling the Kubernetes API, KubeEye gets diagnostic data as it matches key error messages in logs with rules of container syntax on a regular basis. See the architecture graph below:



Here is a list including built-in items that will be checked by KubeEye. Unmarked items are still in development.

YesETCDHealthStatusIf etcd is up and running
YesControllerManagerHealthStatusIf Kubernetes kube-controller-manager is up and running
YesSchedulerHealthStatusIf Kubernetes kube-scheduler is up and running
YesNodeMemoryIf node memory usage is above the threshold
YesDockerHealthStatusIf Docker is up and running
YesNodeDiskIf node disk usage is above the threshold
YesKubeletHealthStatusIf kubelet is active and running normally
YesNodeCPUIf node cpu usage is above the threshold
YesNodeCorruptOverlay2Overlay2 is not available
YesNodeKernelNULLPointerThe node displays NotReady
YesNodeDeadlockA deadlock happens when two or more processes are waiting for each other as they compete for resources
YesNodeOOMMonitor processes that consume too much memory, especially those consuming plenty of memory very quickly. The kernel kills them to prevent them from running out of memory
YesNodeExt4ErrorExt4 mounting error
YesNodeTaskHungCheck if there is a process in state D for more than 120s
YesNodeUnregisterNetDeviceCheck the corresponding network
YesNodeCorruptDockerImageCheck Docker images
YesNodeAUFSUmountHungCheck storage
YesNodeDockerHungAs Docker hangs, you can check Docker logs
YesPodSetLivenessProbeIf the liveness probe is set for every container in a Pod
YesPodSetTagNotSpecifiedThe mirror address does not declare a tag or the tag is latest
YesPodSetRunAsPrivilegedRunning a Pod in a privileged mode means that the Pod can access the hostโ€™s resources and kernel capabilities
YesPodSetImagePullBackOffThe Pod cannot pull the image properly, so it can be pulled manually on the corresponding node
YesPodSetImageRegistryCheck if the image form is in the corresponding registry
YesPodSetCpuLimitsMissingNo CPU resource limit is declared
YesPodNoSuchFileOrDirectoryCheck if the corresponding file exists inside the container
YesPodIOErrorCaused by file IO performance bottlenecks
YesPodNoSuchDeviceOrAddressCheck the corresponding network
YesPodInvalidArgumentCheck the corresponding storage
YesPodDeviceOrResourceBusyCheck the corresponding directory and PID
YesPodFileExistsCheck existing files
YesPodTooManyOpenFilesThe number of open files or socket connections exceeds the system set value
YesPodNoSpaceLeftOnDeviceCheck disk and inode usage
YesNodeApiServerExpiredPeriodCheck if the apiserver certificate expires in less than 30 days
YesPodSetCpuRequestsMissingNo CPU resource request is declared
YesPodSetHostIPCSetSet the host IP
YesPodSetHostNetworkSetSet the host network
YesPodHostPIDSetSet the host PID
YesPodMemoryRequestsMissNo memory resource request is declared
YesPodSetHostPortSet the host port
YesPodSetMemoryLimitsMissingNo memory resource limit is declared
YesPodNotReadOnlyRootFilesThe file system is not set to read-only
YesPodSetPullPolicyNotAlwaysThe image pulling policy is not Always
YesPodSetRunAsRootAllowedExecuted as the root user
YesPodDangerousCapabilitiesYou have the dangerous option in capabilities such as ALL, SYS_ADMIN and NET_ADMIN
YesPodlivenessProbeMissingThe readiness probe is not set
YesprivilegeEscalationAllowedPrivilege escalation is allowed
NodeNotReadyFail to start ContainerManager; cannot set the property TasksAccounting, or unknown properties

Get Started with KubeEye

  1. Clone the KubeEye GitHub repository and install the tool.

    git clone https://github.com/kubesphere/kubeeye.git
    cd kubeeye
    make install


    You can also download the executable file from the GitHub release page of KubeEye.
  2. Install Node Problem Detector (Optional), which will run as a DaemonSet in your cluster. You only need to install NPD when you want detailed reports.

    ke install npd
  3. Use KubeEye to make a diagnosis of your cluster.

    $ ke diag
    NODENAME        SEVERITY     HEARTBEATTIME               REASON              MESSAGE
    node18          Fatal        2020-11-19T10:32:03+08:00   NodeStatusUnknown   Kubelet stopped posting node status.
    node19          Fatal        2020-11-19T10:31:37+08:00   NodeStatusUnknown   Kubelet stopped posting node status.
    node2           Fatal        2020-11-19T10:31:14+08:00   NodeStatusUnknown   Kubelet stopped posting node status.
    node3           Fatal        2020-11-27T17:36:53+08:00   KubeletNotReady     Container runtime not ready: RuntimeReady=false reason:DockerDaemonNotReady message:docker: failed to get docker version: Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?
    NAME            SEVERITY     TIME                        MESSAGE
    scheduler       Fatal        2020-11-27T17:09:59+08:00   Get dial tcp connect: connection refused
    etcd-0          Fatal        2020-11-27T17:56:37+08:00   Get dial tcp connect: connection refused
    NAMESPACE       SEVERITY     PODNAME                                          EVENTTIME                   REASON                MESSAGE
    default         Warning      node3.164b53d23ea79fc7                           2020-11-27T17:37:34+08:00   ContainerGCFailed     rpc error: code = Unknown desc = Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?
    default         Warning      node3.164b553ca5740aae                           2020-11-27T18:03:31+08:00   FreeDiskSpaceFailed   failed to garbage collect required amount of images. Wanted to free 5399374233 bytes, but freed 416077545 bytes
    default         Warning      nginx-b8ffcf679-q4n9v.16491643e6b68cd7           2020-11-27T17:09:24+08:00   Failed                Error: ImagePullBackOff
    default         Warning      node3.164b5861e041a60e                           2020-11-27T19:01:09+08:00   SystemOOM             System OOM encountered, victim process: stress, pid: 16713
    default         Warning      node3.164b58660f8d4590                           2020-11-27T19:01:27+08:00   OOMKilling            Out of memory: Kill process 16711 (stress) score 205 or sacrifice child Killed process 16711 (stress), UID 0, total-vm:826516kB, anon-rss:819296kB, file-rss:0kB, shmem-rss:0kB
    insights-agent  Warning      workloads-1606467120.164b519ca8c67416            2020-11-27T16:57:05+08:00   DeadlineExceeded      Job was active longer than specified deadline
    kube-system     Warning      calico-node-zvl9t.164b3dc50580845d               2020-11-27T17:09:35+08:00   DNSConfigForming      Nameserver limits were exceeded, some nameservers have been omitted, the applied nameserver line is:
    kube-system     Warning      kube-proxy-4bnn7.164b3dc4f4c4125d                2020-11-27T17:09:09+08:00   DNSConfigForming      Nameserver limits were exceeded, some nameservers have been omitted, the applied nameserver line is:
    kube-system     Warning      nodelocaldns-2zbhh.164b3dc4f42d358b              2020-11-27T17:09:14+08:00   DNSConfigForming      Nameserver limits were exceeded, some nameservers have been omitted, the applied nameserver line is:
    NAMESPACE       SEVERITY     NAME                      KIND         TIME                        MESSAGE
    kube-system     Warning      node-problem-detector     DaemonSet    2020-11-27T17:09:59+08:00   [livenessProbeMissing runAsPrivileged]
    kube-system     Warning      calico-node               DaemonSet    2020-11-27T17:09:59+08:00   [runAsPrivileged cpuLimitsMissing]
    kube-system     Warning      nodelocaldns              DaemonSet    2020-11-27T17:09:59+08:00   [cpuLimitsMissing runAsPrivileged]
    default         Warning      nginx                     Deployment   2020-11-27T17:09:59+08:00   [cpuLimitsMissing livenessProbeMissing tagNotSpecified]
    insights-agent  Warning      workloads                 CronJob      2020-11-27T17:09:59+08:00   [livenessProbeMissing]
    insights-agent  Warning      cronjob-executor          Job          2020-11-27T17:09:59+08:00   [livenessProbeMissing]
    kube-system     Warning      calico-kube-controllers   Deployment   2020-11-27T17:09:59+08:00   [cpuLimitsMissing livenessProbeMissing]
    kube-system     Warning      coredns                   Deployment   2020-11-27T17:09:59+08:00   [cpuLimitsMissing]   

Add Custom Check Rules

In addition to the pre-defined diagnostic items and rules, KubeEye also supports customized rules. For example:

  1. Make sure you have installed NPD.

    ke install npd
  2. Edit its ConfigMap using Kubectl.

    kubectl edit cm -n kube-system node-problem-detector-config
  3. Add regular expression rules in the ConfigMap for exceptional logs.

Customize Best Practice Rules

  1. Create a rule in YAML. For example, the rule below checks your Pod specification to make sure images only come from the authorized registry.

      imageFromUnauthorizedRegistry: warning
        promptMessage: When the corresponding rule does not match, show that image from an unauthorized registry.
        category: Images
        target: Container
          '$schema': http://json-schema.org/draft-07/schema
          type: object
              type: string
                pattern: ^quay.io
  2. Save the file, such as rule.yaml.

  3. Run KubeEye with the YAML file.

    $ ke diag -f rule.yaml --kubeconfig ~/.kube/config
    NAMESPACE     SEVERITY    NAME                      KIND         TIME                        MESSAGE
    default       Warning     nginx                     Deployment   2020-11-27T17:18:31+08:00   [imageFromUnauthorizedRegistry]
    kube-system   Warning     node-problem-detector     DaemonSet    2020-11-27T17:18:31+08:00   [livenessProbeMissing runAsPrivileged]
    kube-system   Warning     calico-node               DaemonSet    2020-11-27T17:18:31+08:00   [cpuLimitsMissing runAsPrivileged]
    kube-system   Warning     calico-kube-controllers   Deployment   2020-11-27T17:18:31+08:00   [cpuLimitsMissing livenessProbeMissing]
    kube-system   Warning     nodelocaldns              DaemonSet    2020-11-27T17:18:31+08:00   [runAsPrivileged cpuLimitsMissing]
    default       Warning     nginx                     Deployment   2020-11-27T17:18:31+08:00   [livenessProbeMissing cpuLimitsMissing]
    kube-system   Warning     coredns                   Deployment   2020-11-27T17:18:31+08:00   [cpuLimitsMissing]


Going forward, KubeEye will add support for:

  • Fine-grained diagnostic items, such as the slow response time of Kubernetes clusters.

  • The diagnosis of KubeSphere components, such as DevOps and service mesh.

  • The diagnosis of cluster network and storage.


KubeEye GitHub repository

KubeEye FAQ

Node Problem Detector


Receive the latest news, articles and updates from KubeSphere