{{ $t('productDocDetail.updateTime') }}: 2025-12-18

Backup:

CDP backup data comprises restore point (RP) log files and backup point (BP) backup files. The backup process is primarily based on the generation of BPs and RPs.

1. During VM startup, both the business (QEMU) and CDP (cdp_worker) are enabled. The shared cache is allocated for them on the VM, with a default size of 2 GB.
2. cdp_worker is located at the QCOW2 file layer of the VM and can directly issue backup instructions for the QCOW2 file to generate full and incremental BPs. The initial enabling of CDP requires a full backup for BP generation, and subsequent BPs are generated periodically based on the configured periodic backup frequency (hour-level). The BPs are stored as backup files in the CDP data store.
3. Using a bypass architecture and I/O traffic distribution technology, I/O data is asynchronously copied from the main business to the CDP data store (I/O repository) through the shared cache and exists in the form of I/O logs. RPs are generated periodically based on the configured I/O log interval (the RPO is 1s/5s). However, RPs alone are not enough for data restoration. All RPs depend on their corresponding BPs.
4. The generated BPs and RPs are all timestamped to facilitate RP locating in case of failure.

Backup Consistency:

1. CDP storage has three disks, where an RP with an ID is generated for each I/O operation. RPs that have the same ID on these three disks are considered consistent RPs.
2. For example, in the following figure, the RP whose ID is 3 exists on all three vdisks. As a result, RP3 is a valid consistent RP that can be displayed on the page for VM data restoration.
3. However, the RP whose ID is 6 exists only on vdisk 1 and vdisk 3. As a result, RP6 is an invalid consistent RP that cannot be displayed on the page for VM data restoration.

Restoration:

In the scenario where a business failure occurs (such as virus attack and deletion by mistake) and CDP backup files are required for VM data restoration, the backup files can be found according to the timestamps. Through incremental backups and innovative design of I/O repositories, quick browsing of backup files is supported for file-level and business-level data restoration.

As shown in the above figure, CDP backup files include three BP backup files (BP, BP1, and BP2) and eight RP log files (RP1 to RP8). When a specific RP is selected for restoration, a data merge is performed based on all BPs before the target RP and all RPs between the most recent BP and the target RP for business data restoration.

For example, RP5 is selected for restoration. The following describes the key steps:

1. The endpoint page sends a request to browse the I/O log (RP5) generated at 14:00:20.
2. All backup files before RP5 are extracted from the QCOW2 backup files, that is, the full BP and incremental BP1.
3. All RPs between the most recent BP and the target RP are extracted from the I/O repository, that is, RP4 and RP5.
4. Relevant backup files and RP files obtained are merged to obtain the final result for that BP.
5. You can restore specified documents or data for the entire VM at that moment by retrieving files.
6. In addition, if Sangfor CDP fails, business operations are not affected.

CDP backup and restore in the Kingdee ERP system are taken as an example.

1. The production baseline backup is performed during the initial full backup. Each I/O operation is recorded for RP generation, and incremental BPs are generated periodically.
2. CDP fails after the generation of RP4. However, the business operation is normal.
3. CDP is normal after the generation of RP6, and a BP (incremental BP2) is created immediately to ensure the accuracy of subsequent backup data.

RP5 and RP6 are lost at this time, but other RPs remain intact and valid.