ACK clusters integrate Managed Service for Prometheus to provide visual Prometheus monitoring dashboards. After you enable the dynamic resource overcommitment feature in a cluster, you can use the colocation monitoring feature provided by ACK to view colocation resource metrics, such as the total amount of colocation CPU and memory resources, and the CPU and memory resource allocation rates.
Prerequisites
-
Make sure that the ack-koordinator add-on version is v1.1.1-ack.1 or later. For more information, see ack-koordinator (FKA ack-slo-manager).
-
Make sure that the dynamic resource overcommitment feature is enabled. For more information, see Enable dynamic resource overcommitment.
Dashboard entry
-
Log on to the ACK console. In the left navigation pane, click Clusters.
-
On the Clusters page, click the name of your cluster. In the left navigation pane, click .
-
On the Prometheus Monitoring page, click .
Dashboard overview
The colocation monitoring dashboard provides the following features:
-
Colocation gains: provides a resource gains dashboard for colocation scenarios, where you can view and analyze the resource gains from colocation.
-
Colocation resource observability: provides observability of key metrics in colocation scenarios, where you can query the capacity of colocation resources at the cluster, node pool, node, and pod dimensions.
The following is an example of the colocation monitoring dashboard. The dashboard is upgraded from time to time. Refer to your actual interface.
Cluster colocation gains overview
This area displays the resource gains and usage trends of colocation.
Total and allocated colocation resources
|
Concept |
Description |
|
Non-colocation resources |
All Schedulable (Allocatable) physical resources on ACK nodes are called non-colocation resources. The total amount of non-colocation resources depends on the node specifications and is not affected by colocation. |
|
Colocation resources |
Enabling the dynamic resource overcommitment feature discovers idle physical resources in the cluster, which are the available colocation resources. The total amount of colocation resources depends on the actual resource utilization and changes dynamically with the amount of idle physical resources on nodes. Colocation resources are additional Schedulable resources provided by colocation and are one of the key metrics for measuring colocation gains. |
|
Total colocation resources |
Divided into two resource dimensions: CPU and memory, which display the total amount of CPU resources and the total amount of memory resources currently available for colocation scheduling in the cluster. The cluster has a total of 118 CPU cores and 487 GiB of memory resources available for colocation. A larger amount of total colocation resources indicates more idle physical resources available for colocation in the cluster and greater potential gains from scheduling more applications through colocation. |
|
Allocated colocation resources |
Divided into two resource dimensions: CPU and memory, which display the amount of CPU resources and the amount of memory resources currently allocated for colocation in the cluster. A total of 2 CPU cores and 1 GiB of memory resources have been allocated for colocation in the cluster. A larger amount of allocated colocation resources indicates more colocation resources scheduled in the cluster and greater gains from scheduling more applications through colocation. |
|
Colocation resource allocation rate |
Divided into two resource dimensions: CPU and memory, which display the CPU resource allocation rate and the memory resource allocation rate for colocation in the cluster ( A higher colocation resource allocation rate indicates a higher proportion of colocation resources allocated in the cluster and greater colocation gains. |
Colocation resource usage trends
|
Concept |
Description |
|
Number of colocation pods |
Divided into the number of pods scheduled with non-colocation resources and the number of pods scheduled with colocation resources. The colocation pod ratio displays the relative proportion of pods scheduled with non-colocation resources and pods scheduled with colocation resources in the cluster. |
|
Colocation resource ratio |
Divided into two resource dimensions: CPU and memory. The colocation resource ratio displays the amount of resources currently available for non-colocation scheduling and the amount of resources available for colocation scheduling in the cluster. The more idle physical resources in the cluster, the higher the proportion of colocation resources, which indicates that more resources are available for colocation scheduling. |
Cluster colocation resource details
Cluster resource view, Node resource view, and Pod resource view areas display resource usage and resource requests at the cluster, node, and pod dimensions, respectively.
Cluster resource view
|
Concept |
Description |
|
Cluster resource usage |
Divided into two resource dimensions: CPU and memory. Each dimension includes the total physical resources of the cluster, the resource usage of non-colocation pods, the resource usage of colocation pods, and the resource usage of system components. Cluster resource usage measures the usage of different types of resources in the current cluster. If the sum of the three usage metrics is much smaller than the total physical resources, the average utilization of the cluster is low and many physical resources are idle. |
|
Cluster colocation resource requests |
Divided into two resource dimensions: CPU and memory. Each dimension includes the total amount of Schedulable colocation resources and the amount of colocation resources currently requested. The closer the resource requests are to the total resources, the higher the proportion of colocation resources requested, which indicates that colocation resources are more fully allocated. |
|
Cluster non-colocation resource requests |
Divided into two resource dimensions: CPU and memory. Each dimension includes the total amount of Schedulable non-colocation resources and the amount of non-colocation resources currently requested. The closer the resource requests are to the total resources, the higher the proportion of non-colocation resources requested, which indicates that non-colocation resources are more fully allocated. |
Node resource view
In the upper part of the page, configure node_label and node_label_value to view the node resource view of different nodes.
The filter bar also contains the Node, pod_namespace, and pod_name filters, which allow you to further refine the data scope of the node resource view by node, namespace, and pod name.
The following table describes some of the filters:
|
Filter |
Description |
|
node_label_value |
The default value is All, which indicates that the statistical scope of the cluster colocation gains and cluster resource view areas covers all nodes in the cluster. You can select a target node pool to view the resource view of the node pool in the cluster colocation gains and cluster resource view areas. |
|
node_label |
Supports selecting nodes by different node labels. For more information, see Note at the top of the page. |
|
Concept |
Description |
|
Node colocation resource ratio |
Divided into two resource dimensions: CPU and memory. Each dimension includes the total amount of non-colocation resources and the total amount of colocation resources on the node. The total non-colocation resources and the total colocation resources are displayed in a stacked manner, which indicates the relative proportion of the two in the node resources. |
|
Node resource usage |
Divided into two resource dimensions: CPU and memory. Memory resources are further divided into two sub-dimensions, excluding cache and including cache, based on whether the cache is counted. Each dimension includes the total physical resources on the node, the resource usage of non-colocation pods, the resource usage of colocation pods, and the resource usage of system components. Node resource usage measures the usage of different types of resources on the current node. If the sum of the three usage metrics is much smaller than the total physical resources, the average utilization of the node is low and many physical resources are idle. |
|
Node colocation resource requests |
Divided into two resource dimensions: CPU and memory. Each dimension includes the total amount of colocation resources and the amount of colocation resources requested on the node. The closer the resource requests are to the total resources, the higher the proportion of colocation resources requested, which indicates that colocation resources are more fully allocated. |
|
Colocation resource requests of each pod |
Divided into two resource dimensions: CPU and memory. Includes the resource requests of all colocation pods on the node. |
|
Colocation resource utilization of each pod |
Divided into two resource dimensions: CPU and memory. Includes the resource utilization of all colocation pods on the node. |
Pod resource view
Above the k8s-reclaimed-resource tab, configure pod_namespace and pod_name to view the resource view of different pods.
|
Concept |
Description |
|
Pod colocation resources |
Divided into two resource dimensions: CPU and memory. Each dimension includes the upper limit, requests, and actual usage of colocation resources for the pod. |
|
Pod colocation resource utilization |
Divided into two resource dimensions: CPU and memory. Each dimension includes the utilization of colocation resources for the pod. |
|
Colocation resources of each container |
Divided into two resource dimensions: CPU and memory. Each dimension includes the upper limit, requests, and actual usage of colocation resources for each container in the pod. |
FAQ
Why is no data displayed in the cluster colocation gains area on the Kubernetes Reclaimed Resource tab?
-
Check whether the ack-koordinator add-on is installed.
-
Log on to the ACK console. In the left navigation pane, click Clusters.
-
On the Clusters page, click the name of your cluster. In the left navigation pane, click Components and Add-ons .
-
Search for and check whether the ack-koordinator add-on is displayed as installed.
For component versions earlier than v0.7 deployed from Marketplace, see Migrate ack-koordinator from Marketplace to add-ons to complete the migration.
-
Not installed: Follow the on-screen instructions to complete the installation.
-
Installed: Proceed with the following steps.
-
-
-
Check whether the colocation monitoring dashboard displays relevant data.
If no data is displayed, perform the following steps:
Log on to the ARMS console.
-
In the left navigation pane, choose .
-
In the upper-left corner of the page, select the target region and click the name of the Prometheus instance. Then, in the left-side navigation pane, click Metric Management.
-
In the Metrics exploration area, search for and select kube_node_labels, and view the data details of the metric as prompted.