Constraints

Clusters and Nodes

• After a cluster is created, the following items cannot be changed:
  • Number of master nodes: For example, a non-HA cluster (with one master node) cannot be changed to an HA cluster (with three master nodes).
  • AZ where a master node is deployed
  • Network configuration of the cluster, such as the VPC, subnet, container CIDR block, Service CIDR block, IPv6 settings, and kube-proxy (forwarding) settings.
  • Network model: For example, a container tunnel network cannot be changed to a VPC network.
• CCE underlying resources such as ECS nodes are limited by quota and their inventory. It is possible that only some nodes are created during cluster creation, cluster scaling, or auto scaling.
• ECS node specifications: CPU ≥ 2 cores, memory ≥ 4 GiB
• To access a CCE cluster through a VPN, ensure that the VPN CIDR block does not conflict with the VPC CIDR block where the cluster resides and the container CIDR block.

Networks

• By default, a NodePort Service is accessed within a VPC. To access a NodePort Service through the Internet, bind an EIP to the node in the cluster beforehand.
• LoadBalancer Services allow workloads to be accessed from public networks through ELB. This access mode has the following restrictions:
  • Automatically created load balancers should not be used by other resources. Otherwise, these load balancers cannot be completely deleted.
  • Do not change the listener name for the load balancer in clusters of v1.15 and earlier. Otherwise, the load balancer cannot be accessed.
• Constraints on network policies:
  • Only clusters that use the tunnel network model support network policies. Network policies are classified into the following types:
    • Ingress: All versions support this type.
    • Egress: This rule type cannot be set currently.
  • Network isolation is not supported for IPv6 addresses.

Storage Volumes

• Constraints on EVS volumes:
  • EVS disks cannot be attached across AZs and cannot be used by multiple workloads, multiple pods of the same workload, or multiple tasks. Data sharing of a shared disk is not supported between nodes in a CCE cluster. If an EVS disk is attacked to multiple nodes, I/O conflicts and data cache conflicts may occur. Therefore, create only one pod when creating a Deployment that uses EVS disks.
  • For clusters earlier than v1.19.10, if an HPA policy is used to scale out a workload with EVS disks attached, the existing pods cannot be read or written when a new pod is scheduled to another node.

    For clusters of v1.19.10 and later, if an HPA policy is used to scale out a workload with EVS disks attached, a new pod cannot be started because EVS disks cannot be attached.

• Constraints on OBS volumes:
  • When parallel file systems and object buckets are used, the group and permission of the mount point cannot be modified.
  • CCE allows you to use OBS parallel file systems by calling the OBS SDK or mounting a PVC through the obsfs tool provided by OBS. Each time an OBS parallel file system is mounted, an obsfs resident process is generated. The following shows an example.
    Figure 1 obsfs resident process
    

    Reserve 1 GiB of memory for each obsfs process. For example, for a node with 4 vCPUs and 8 GiB of memory, an obsfs parallel file system should be mounted to no more than eight pods.

    

    An obsfs resident process runs on a node. If the consumed memory exceeds the upper limit of the node, the node malfunctions. On a node with 4 vCPUs and 8 GiB of memory, if more than 100 pods are mounted to a parallel file system, the node will be unavailable. Control the number of pods mounted to a parallel file system on a single node.

• Constraints on local PVs:
  • Local PVs are supported only when the cluster version is v1.21.2-r0 or later and the everest add-on version is 2.1.23 or later. Version 2.1.23 or later is recommended.
  • Deleting, removing, resetting, or scaling in a node will cause the PVC/PV data of the local PV associated with the node to be lost, which cannot be restored or used again. In these scenarios, the pod that uses the local PV is evicted from the node. A new pod will be created and stay in the pending state. This is because the PVC used by the pod has a node label, due to which the pod cannot be scheduled. After the node is reset, the pod may be scheduled to the reset node. In this case, the pod remains in the creating state because the underlying logical volume corresponding to the PVC does not exist.
  • Do not manually delete the corresponding storage pool or detach data disks from the node. Otherwise, exceptions such as data loss may occur.
  • A local PV cannot be mounted to multiple workloads or jobs concurrently.
• Constraints on local EVs:
  • Local EVs are supported only when the cluster version is v1.21.2-r0 or later and the Everest add-on version is 1.2.29 or later.
  • Do not manually delete the corresponding storage pool or detach data disks from the node. Otherwise, exceptions such as data loss may occur.
  • Ensure that the /var/lib/kubelet/pods/ directory is not mounted to the pod on the node. Otherwise, the pod, mounted with such volumes, may fail to be deleted.
• Constraints on snapshots and backups:
  • The snapshot function is available only for clusters of v1.15 or later and requires the CSI-based everest add-on.
  • The subtype (common I/O, high I/O, or ultra-high I/O), disk mode (SCSI or VBD), sharing status, and capacity of an EVS disk created from a snapshot must be the same as those of the disk associated with the snapshot. These attributes cannot be modified after being queried or set.
  • Snapshots can be created only for EVS disks that are available or in use, and a maximum of seven snapshots can be created for a single EVS disk.
  • Snapshots can be created only for PVCs created using the storage class (whose name starts with csi) provided by the everest add-on. Snapshots cannot be created for PVCs created using the Flexvolume storage class whose name is ssd, sas, or sata.

Add-ons

CCE uses Helm charts to deploy add-ons. To modify or upgrade an add-on, perform operations on the Add-ons page or use open add-on management APIs. Do not directly modify add-on resources on the backend. Otherwise, add-on exceptions or other unexpected problems may occur.

CCE Cluster Resources

There are resource quotas for your CCE clusters in each region.

Dependent Underlying Cloud Resources