Notes on Using the HCI Cluster Shrink Tool

No.

Issue Description

Remarks

1

Compound datastores cannot be shrunk.

2

The datastore contains at least four data nodes.

3

The node to be removed must in running state, and the storage area network and management network of HCI function normally.

4

If remove the host fails and it is confirmed that there are ongoing tasks in the cluster, wait for these tasks to complete before retrying. If multiple retry attempts fail, contact Sangfor technical support for assistance.

5

You can only remove one node at a time.

6

During the shrinking process, you cannot perform any O&M task on the cluster, such as replacement, expansion, maintenance, or slow disk processing. When an O&M task is in progress, you cannot remove a node.

7

Shrinking a virtual datastore from 3 replicas to 2 replicas is not supported.

8

HCI 6.3.0 does not support shrinking a cluster from 5 nodes to 4 nodes.

9

Shrinking is not supported when exceptions occurred on replicas.

10

After you have manually exited maintenance mode, you must re-enter maintenance mode by selecting the Prioritize data security mode to proceed with the shrinking process.

11

HCI Cluster Shrink can be used when the following condition  is met: the sum of the storage used by the shrinked host (Host A) and the storage used by any other host in the cluster (Host B), divided by the total storage capacity of Host B, must be less    than 90%.

12

Currently only supports: HCI630R3, HCI670R3, HCI680, HCI680R1, HCI680R2, HCI681, HCI681R1, HCI690, HCI691,    including their corresponding Chinese and International       versions, as well as beta versions.

13

During the shrinking process, the tool automatically disables data balancing. Once the data migration is complete, data balancing is enabled again when you click Continue. You can also manually enable data balancing.

14

If the disk group data migration fails and you retry, the progress will restart from 0%.

The migrated data will not be migrated again.

15

For HCI 6.3.0_R3, if the execution host (VS master controller) is restarted, and for the 6.7.0_R3 and 6.8.0 series, if the host being shrunk is restarted during the shrink process, the shrink operation will get stuck. If you encounter a stuck shrink operation, please contact a Sangfor technical support.

Do not restart any node during the shrinking process.

16

During the shrinking process, performance may degrade for one to two seconds multiple times. You are advised to perform shrinking during off-peak hours.

17

On HCI of a version earlier than 6.7.0, if no more than 20 shared disks are connected to the node to be removed, the business traffic will be interrupted for one to two seconds. If the number exceeds 20, contact a Sangfor technical support representative for evaluation.

Evaluation is necessary only if the total number of shared disks divided by the total number of nodes in the datastore is greater than 20.

18

To use the HCI Cluster Shrink tool, you need to log in to the HCI cluster controller. If you switch the cluster controller after logging in, the remove of the host will fail. You will need to log in again to successfully remove the host.

To remove the cluster controller of a cluster, first switch the role of the cluster controller to another node, and then use the HCI Cluster Shrink tool of aDeploy to remove it.

19

If the storage usage will exceed 80% after the shrinking process, the shrinking operation is blocked during the pre-shrink check.

20

Exceptions or interruptions of aDeploy do not affect the shrinking process. In this case, start aDeploy again. If aDeploy cannot be started, contact a technical support representative.

21

Node removal requires node data rebuilding, which takes 3 hours per TB.

Please perform shrinking during idle hours of a node.

22

To downsize a 5-host cluster, at least 3 of the remaining hosts (excluding the downsized host) must have 2 or more disk groups.

23

The 6.10.0 series multi-volume shrinkage has known issues, and the latest cumulative patch package must be applied before performing shrinkage

24

There is a significant impact on business during the shrinkage process. Under stress conditions of reading at 456MB/s and writing at 197MB/s while shrinking one host, business performance may decline by 50-90% during the shrinkage period, and there may be one instance of a 0 drop lasting an average of 1-2 seconds.