Feature Overview
Component Description
Node-Problem-Detector-Plus is an add-on that monitors the health status of Kubernetes cluster nodes. It runs in the TKE environment as a DaemonSet to help users detect various exceptions on nodes in real time and report the detection results to the upstream Kube-apiserver.
Kubernetes objects deployed within a cluster
Kubernetes Object Name | Local Disk Types | Resource Amount | Namespaces |
node-problem-detector | DaemonSet | 0.5C80M | kube-system |
node-problem-detector | ServiceAccount | - | kube-system |
node-problem-detector | ClusterRole | - | - |
node-problem-detector | ClusterRoleBinding | - | - |
Use Cases
The Node-Problem-Detector-Plus add-on enables monitoring of node working status, including kernel deadlock, OOM, system thread count pressure, and system file descriptor pressure. These metrics are reported to the Apiserver in the form of Node Condition and Event.
By monitoring these metrics, you can anticipate resource pressure on nodes and manually release or scale up node resources before the node starts evicting Pods. This helps prevent potential losses caused by Kubernetes resource reclamation or node unavailability.
Limits
To use NPD in your cluster, you need to install this add-on in your cluster. The system resources used by NPD containers is restricted to 0.5 CPU core and 80 MB memory.
How to Use
1. Log in to the TKE console and select Cluster in the left sidebar.
2. On the "Cluster Management" page, click the ID of the target cluster to enter the cluster details page.
3. Select Component Management from the left sidebar menu to access the "Component List" page.
4. In the "Component List" page, select Create and check Node-Problem-Detector-Plus in the "Create Component" page.
5. Click Done to create the component. After successful installation, the corresponding node-problem-detector resources will be available in your cluster, and additional entries will be added to the Node's Condition.
See Also
Node Conditions
After the NPD plug-in is installed, the following specific Conditions will be added to nodes:
Condition Type | Default value | Description |
ReadonlyFilesystem | False | Indicate whether the file system is read-only |
FDPressure | False | Query whether the number of file descriptors of the host reaches 80% of the maximum value |
FrequentKubeletRestart | False | Indicates whether Kubelet has restarted more than 5 times in 20 minutes. |
CorruptDockerOverlay2 | False | Indicate whether DockerImage is faulty |
KubeletProblem | False | Indicates whether the Kubelet service is Running. |
KernelDeadlock | False | Indicate whether a deadlock occurs in the kernel |
FrequentDockerRestart | False | Indicates whether Docker has restarted more than 5 times in 20 minutes. |
FrequentContainerdRestart | False | Indicates whether Containerd has restarted more than 5 times in 20 minutes. |
DockerdProblem | False | Indicates whether the Docker service is Running (if the node runtime is Containerd, the value is always False). |
ContainerdProblem | False | Indicates whether the Containerd service is Running (if the node runtime is Docker, the value is always False). |
ThreadPressure | False | Indicate whether the current number of threads of the system reaches 90% of the maximum value |
NetworkUnavailable | False | Indicates whether the NTP service status is Running. |
SerfFailed | False | Detect the node network health in distributed mode |