Repair lives on the instance detail page, in the Advanced panel. Before you select it, confirm the instance has no data that is both unbacked up and irreplaceable.
Repair is for underlying resource failures or an instance that cannot start normally — for example, the host resources behind the virtual machine are damaged, or the system disk fails to boot and a restart does not fix it. The dashboard describes it this way:
Repair the instance with its currently saved image, shape, network, and login configuration when underlying resources fail or the instance cannot start normally.
If the instance is merely misbehaving but can still power on and off, try a restart first, not repair. A restart preserves disk data; repair does not.
The name "repair" sounds like a patch. What it actually does is delete the virtual machine and rebuild it from scratch:
| What happens | What does not happen |
|---|---|
| If the instance is running, the platform force-stops it first, then destroys the virtual machine | The old virtual machine's disk data is not preserved in any form — it is a full wipe, not a reformat that keeps the same disk |
| Every storage volume this instance owns at the virtualization layer is permanently deleted | The instance's identity, network configuration, and billing relationship are unchanged, and existing backup records are unaffected |
| A new virtual machine is created from a template, using the instance's currently saved image, shape, network, and login configuration | The two-step confirmation is not skippable — both the frontend and the backend validate that the name you typed matches the instance |
| The instance state changes to Repairing and it is temporarily offline | Repair does not fail partway through because a dependency turned out to be unavailable — shape, image, and node configuration are all re-validated before anything is deleted, and any failure there blocks the operation before it touches the existing virtual machine |
The current instance model has a single system disk with no separate data-disk concept, so "wipes all disk data" means wiping that one system disk. Repair does not take a snapshot for you automatically — if you already have a full-instance backup, you can recover the data from it; if you do not, the data cannot be recovered.
- Permissions: your role must include the instance operate permission. Project members and above have it by default; project viewers do not, and the button stays disabled.
- The instance state must allow repair: for a provisioned instance, only Running, Stopped, Error, and Failed allow repair. For an instance that has never finished provisioning, only Pending, Provisioning, Error, and Failed allow it. In any other state, the button stays disabled.
- The instance must not be under maintenance lock: repair is unavailable while the host is in maintenance.
- No work request in progress: repair is rejected while a create, rebuild, upgrade, migrate, or delete task is pending or running.
-
Go to Compute > Instances, open the target instance's detail page, and switch to the Advanced panel.
-
In the Repair instance block, review the three impact rows: Keep — "Instance identity, networking, billing relationship, and backup records"; Delete — "All data on the current virtual machine disks, with no recovery option"; Impact — "The instance will be temporarily offline while it is repaired".
-
Select Repair instance to open the confirmation dialog.
-
The dialog asks you to type the instance name to confirm:
Type the instance name "
{{name}}" to confirm clearing the current disks and repairing the instance:{{name}}is replaced with this instance's actual name. Type the current instance name exactly into the field.Interface illustration：Repair instance confirmation dialog: title, the Keep / Delete / Impact list, the instructions to type the instance name, the name input field, and the Cancel and confirm buttons at the bottom -
If what you typed does not match the instance name, the field shows "The name you entered doesn't match the instance name" and the confirm button stays disabled. Once the name matches exactly, the confirm button lights up as Clear disks and repair instance.
-
Select Clear disks and repair instance to submit.
After you submit, the instance state changes to Repairing, and you can track progress under Work requests on the instance detail page. Once repair finishes, the state returns to Running or Stopped, depending on whether the new virtual machine powers on successfully.
If this instance's original creation failed partway through and the platform never successfully created a virtual machine for it, selecting Repair instance takes a completely different path: the platform resumes the original, unfinished creation request and continues provisioning the instance. No disk is destroyed, because there is no existing virtual machine to delete.
This applies to instances still in Pending, Provisioning, Error, or Failed state that have never successfully powered on. For these instances, "repair" is effectively "resume creation" and is safe to use. The destructive delete-and-recreate behavior only happens once the instance has an actual, existing virtual machine.
Repair is validated before it is submitted. The dashboard displays the platform's message verbatim, and these messages are currently returned in Chinese regardless of your interface language. Each message is quoted below, followed by what to do.
此实例正有工作请求正在进行，因此您无法操作此实例。请等待最长 15 分钟，然后重试。如果您遇到问题，请联系 Cypacky Support 以获取帮助。
A create, rebuild, upgrade, migrate, or delete task on this instance is pending or running. Open Work requests on the instance detail page and wait for the last task to finish before retrying.
此实例正在维护，当前无法修复。
The host the instance runs on is in maintenance. This is a check specific to repair — the wording differs from the maintenance message shown for power actions. Try again once maintenance ends.
实例当前状态不允许修复
The instance is not in a state that permits repair; refer to the state list under "Before you begin". In most cases the button is already disabled before you can reach this message.
输入的名称与实例名称不一致
The text in the confirmation field must match the instance name exactly before the confirm button lights up. Even if you bypass the frontend and call the API directly, the backend re-validates the same rule and returns "实例名称不匹配" ("Instance name mismatch").
实例配置已不可用，无法修复
The shape this instance uses has been retired or is no longer valid. Repair re-validates the shape before deleting anything, and rejects the request outright if it has become unavailable — the existing virtual machine is left untouched.
实例映像已不可用，无法修复
The image version this instance uses has been retired or is no longer valid, so repair has nothing to clone the new virtual machine from and is rejected outright.
节点 PVE 配置不可用，无法修复
The virtualization configuration for the node this instance runs on is unavailable. This is usually a node-side issue, and no deletion happens in this case. Retry later, or contact Cypacky Support.
此实例正在等待支付激活，当前无法执行该操作。
此实例当前处于欠费状态，当前无法执行该操作。
Complete the payment or settle the outstanding balance, then repair becomes available.
此实例有待支付的升级或流量重置账单，当前操作可能与付款兑现冲突。
此实例有已支付账单正在兑现，请等待账单动作完成后再操作。
Complete or cancel the related invoice, or wait for the invoice's changes to finish applying, then retry.
此实例因安全投诉已被锁定。当前操作不可用，请联系 Cypacky Support。
Contact Cypacky Support through a ticket to learn the reason and the conditions for unlocking.