Backup and Restoration

1. Create a path for storing the backup file package on the remote host over SFTP.
2. Run the following backup command:

   ./ucs-ctl backup {Cluster name} --path {Backup path} --type sftp --ip {IP address of the remote host} --user {Username of the remote host}

   Example:

   ./ucs-ctl backup gpu-test --path /home/ggz/gpu-test --type sftp --ip 100.95.142.93 --user root

   

   • If you perform remote backup for the first time, enter the password of the remote host after "please input sftp password:" is displayed.
   • The backup path in the backup command must be valid. Otherwise, the backup file generated on the remote host after the command is executed may be incorrect.

   After the command is executed successfully, a backup file package in the format of {Cluster name}-backup-{Timestamp}.tar.gz is generated in the specified backup path on the remote host. The package stores the ha.yaml file and the etcd-snapshot and crt directories. The etcd-snapshot directory contains etcd data, and the crt directory contains certificates and encryption and decryption materials.

Periodic backup

Run the crontab -e command to compile a crontab expression so that the backup command is executed periodically.

In the following example, the local backup task of the cluster is executed at 16:40 every day.

40 16 * * * /root/cluster/ucs-ctl backup cluster-redhat --path /root/cluster/backup --type local  

During periodic remote backup, you do not need to specify the password in the crontab expression after entering the password of the remote host for the first time.

To prevent the number of backup files from increasing, compile the crontab expression on the remote host to periodically execute the backup file aging script. The following is an example of the backup file aging script:

#!/bin/bash  
backup_dir=${1}                                            # Path for storing backup files
keep_days=${2}                                             # Retention days of backup files
if [ ! -d ""$backup_dir"" ]; then                          # Check whether the log path exists.              
  echo "There is no backup file path."
  exit 1 
fi  
find "$backup_dir" -type f -mtime +$keep_days -exec rm {} \;       # Delete legacy logs.
echo "Expired backup files have been deleted!"

1. Prepare an etcd data backup package in the format of {Cluster name}-backup-{Timestamp}.tar.gz.

   Upload the package to each master/etcd node of the cluster.

2. Stop the etcd service.

   Run the following command on the node:

   mv /var/paas/kubernetes/manifests/etcd*.manifest /var/paas/kubernetes/

   Wait until the service is stopped.

   crictl ps | grep etcd

   If no etcd container is found, the service has been stopped.

   

3. (Optional) Back up the etcd data on a node.

   mv /var/paas/run/etcd/data /var/paas/run/etcd/data-bak 
   mv /var/paas/run/etcd-event/data /var/paas/run/etcd-event/data-bak

4. Run the restoration command on the node where the etcd database resides.

   ./ucs-ctl restore etcd {Path of the etcd data backup package}

   Example:

   ./ucs-ctl restore etcd /home/ggz/gpu-test/backup-file-20230625164904.tar.gz

   The etcd data is restored if the following command output is displayed:

   Restore the etcd snapshot successfully.

5. Restart the etcd service on the node. The restart takes several minutes.

   mv /var/paas/kubernetes/etcd*.manifest /var/paas/kubernetes/manifests

   Wait for the service to restart.

   crictl ps | grep etcd

   If the etcd containers are found, the service has been restarted and the etcd data is restored on the node.

   

   

   To restore etcd data, perform steps 1 to 5 on each node where the etcd database resides.

Restoring a single master node

1. Run the single-node fault recovery command on the executor:

   ./ucs-ctl restore node {IP address of the node} --name {Cluster name}

   {IP address of the node} indicates the IP address of the faulty node. The following is an example:

   ./ucs-ctl restore node 192.168.0.87 --name gpu-test

   The fault on the single master node is rectified if the following command output is displayed:

   restore node 192.168.0.87 successfully.

2. Restore etcd data on the node where etcd is located. For details about the commands, see 1 to 5.

Restoring master nodes

1. Clear the data on all master nodes. Copy the uninstallation script in the executor directory to the master nodes and execute the script on each master node. (The script path is /var/paas/.ucs-package/ucs-onpremise/scripts/uninstall_node.sh.)

   scp -r /var/paas/.ucs-package/ucs-onpremise/scripts/uninstall_node.sh root@{IP address of the target node}:/root
   sudo bash /root/uninstall_node.sh

2. Run the cluster fault recovery command on the executor:

   ./ucs-ctl restore cluster {Cluster name} -b {Path of the backup file package}

   Example command:

   ./ucs-ctl restore cluster gpu-test -b /home/ggz/gpu-test/backup-file-20230625164904.tar.gz

   The faults on the master nodes are rectified if the following command output is displayed:

   restore cluster successfully.

3. Restore etcd data on each node where etcd is located. For details about the commands, see 1 to 5.