TiDB Cloud
TiDB Self-Managed
Learning Center
Contact Us
​
Sign InStart for Free

Backup & Restore FAQ

This document lists the frequently asked questions (FAQs) and the solutions about Backup & Restore (BR).

What should I do if the error message could not read local://...:download sst failed is returned during data restoration?

When you restore data, each node must have access to all backup files (SST files). By default, if local storage is used, you cannot restore data because the backup files are scattered among different nodes. Therefore, you have to copy the backup file of each TiKV node to the other TiKV nodes.

It is recommended to mount an NFS disk as a backup disk during backup. For details, see Back up a single table to a network disk.

How much does it affect the cluster during backup using BR?

For TiDB v5.3.0 or earlier versions, the default configuration parameters of BR only serve to offline backup tasks. In the default configuration, the backup tasks using BR might use a large amount of CPU and I/O resources of the cluster, which increases cluster latency.

The results of internal tests show that under the default configuration, the backup tasks using BR have a great impact on the cluster performance. The detailed test information is as follows:

  • Test environment: Use a cluster with 6 TiKV nodes and a single-table dataset to perform the following hybrid operations. The ratio of the read and write tasks is about 10:1.
  • Test methods:
    • Insert data to a single point
    • Update data in a single point
    • Queries for small ranges
  • Benchmark: Total QPS is around 20k. P95 latency is around 28 ms. P99 latency is around 50 ms.
  • Test result: When the total CPU utilization of the cluster is limited to 91%, the results are as follows:
    • The QPS of write workload reduced by 50%, and P99 latency increased by several times.
    • Read workload is less affected, and its QPS even increased slightly.
    • The overall P99 latency only rose by around 50%, and QPS reduced by around 4%.

If you need to manually control the impact of backup tasks on cluster performance, you can use the following solutions. These two methods can reduce the impact of backup tasks on the cluster, but they also reduce the speed of backup tasks.

  • Use the --ratelimit parameter to limit the speed of backup tasks. Note that this parameter limits the speed of storing backup files to external storage. When calculating the total size of backup files, use the backup data size(after compressed) in the backup log as a benchmark.
  • Adjust the TiKV configuration item backup.num-threads to limit the resources used by backup tasks. This configuration item determines the number of threads used by backup tasks. The test data shows that when the number of threads used by backup tasks using BR does not exceed 8, and the total CPU utilization of the cluster does not exceed 60%, the backup tasks have little impact on the cluster for both read and write workload.

Does BR back up system tables? During data restoration, do they raise conflicts?

Before v5.1.0, BR filtered out data from the system schema mysql during the backup. Since v5.1.0, BR backs up all data by default, including the system schemas mysql.*.

During data restoration, system tables do not raise conflicts. The technical implementation of restoring the system tables in mysql.* is not complete yet, so the tables in the system schema mysql are not restored by default, which means no conflicts will be raised. For more details, refer to the Back up and restore table data in the mysql system schema (experimental feature).

What should I do to handle the Permission denied or No such file or directory error, even if I have tried to run BR using root in vain?

You need to confirm whether TiKV has access to the backup directory. To back up data, confirm whether TiKV has the write permission. To restore data, confirm whether it has the read permission.

During the backup operation, if the storage medium is the local disk or a network file system (NFS), make sure that the user to start BR and the user to start TiKV are consistent (if BR and TiKV are on different machines, the users' UIDs must be consistent). Otherwise, the Permission denied issue might occur.

Running BR with the root access might fail due to the disk permission, because the backup files (SST files) are saved by TiKV.

  1. Run the Linux-native command for process query:

    ps aux | grep tikv-server

    The output of the above command:

    tidb_ouo  9235 10.9  3.8 2019248 622776 ?      Ssl  08:28   1:12 bin/tikv-server --addr 0.0.0.0:20162 --advertise-addr 172.16.6.118:20162 --status-addr 0.0.0.0:20188 --advertise-status-addr 172.16.6.118:20188 --pd 172.16.6.118:2379 --data-dir /home/user1/tidb-data/tikv-20162 --config conf/tikv.toml --log-file /home/user1/tidb-deploy/tikv-20162/log/tikv.log
tidb_ouo  9236  9.8  3.8 2048940 631136 ?      Ssl  08:28   1:05 bin/tikv-server --addr 0.0.0.0:20161 --advertise-addr 172.16.6.118:20161 --status-addr 0.0.0.0:20189 --advertise-status-addr 172.16.6.118:20189 --pd 172.16.6.118:2379 --data-dir /home/user1/tidb-data/tikv-20161 --config conf/tikv.toml --log-file /home/user1/tidb-deploy/tikv-20161/log/tikv.log

    Or you can run the following command:

    ps aux | grep tikv-server | awk '{print $1}'

    The output of the above command:

    tidb_ouo
tidb_ouo

  2. Query the startup information of the cluster using the TiUP command:

    tiup cluster list

    The output of the above command:

    [root@Copy-of-VM-EE-CentOS76-v1 br]# tiup cluster list
Starting component `cluster`: /root/.tiup/components/cluster/v1.5.2/tiup-cluster list
Name          User      Version  Path                                               PrivateKey
----          ----      -------  ----                                               ----------
tidb_cluster  tidb_ouo  v5.0.2   /root/.tiup/storage/cluster/clusters/tidb_cluster  /root/.tiup/storage/cluster/clusters/tidb_cluster/ssh/id_rsa

  3. Check the permission for the backup directory. For example, backup is for backup data storage:

    ls -al backup

    The output of the above command:

    [root@Copy-of-VM-EE-CentOS76-v1 user1]# ls -al backup
total 0
drwxr-xr-x  2 root root   6 Jun 28 17:48 .
drwxr-xr-x 11 root root 310 Jul  4 10:35 ..

    From the above output, you can find that the tikv-server instance is started by the user tidb_ouo. But the user tidb_ouo does not have the write permission for backup, the backup fails.

What should I do to handle the Io(Os...) error?

Almost all of these problems are system call errors that occur when TiKV writes data to the disk. For example, if you encounter error messages such as Io(Os {code: 13, kind: PermissionDenied...}) or Io(Os {code: 2, kind: NotFound...}), you can first check the mounting method and the file system of the backup directory, and try to back up data to another folder or another hard disk.

For example, you might encounter the Code: 22(invalid argument) error when backing up data to the network disk built by samba.

What should I do to handle the rpc error: code = Unavailable desc =... error occurred in BR?

This error might occur when the capacity of the cluster to restore (using BR) is insufficient. You can further confirm the cause by checking the monitoring metrics of this cluster or the TiKV log.

To handle this issue, you can try to scale out the cluster resources, reduce the concurrency during restore, and enable the RATE_LIMIT option.

Where are the backed up files stored when I use local storage?

When you use local storage, backupmeta is generated on the node where BR is running, and backup files are generated on the Leader nodes of each Region.

How about the size of the backup data? Are there replicas of the backup?

During data backup, backup files are generated on the Leader nodes of each Region. The size of the backup is equal to the data size, with no redundant replicas. Therefore, the total data size is approximately the total number of TiKV data divided by the number of replicas.

However, if you want to restore data from local storage, the number of replicas is equal to that of the TiKV nodes, because each TiKV must have access to all backup files.

What should I do when BR restores data to the upstream cluster of TiCDC/Drainer?

  • The data restored using BR cannot be replicated to the downstream. This is because BR directly imports SST files but the downstream cluster currently cannot obtain these files from the upstream.

  • Before v4.0.3, DDL jobs generated during the BR restore might cause unexpected DDL executions in TiCDC/Drainer. Therefore, if you need to perform restore on the upstream cluster of TiCDC/Drainer, add all tables restored using BR to the TiCDC/Drainer block list.

You can use filter.rules to configure the block list for TiCDC and use syncer.ignore-table to configure the block list for Drainer.

Does BR back up the SHARD_ROW_ID_BITS and PRE_SPLIT_REGIONS information of a table? Does the restored table have multiple Regions?

Yes. BR backs up the SHARD_ROW_ID_BITS and PRE_SPLIT_REGIONS information of a table. The data of the restored table is also split into multiple Regions.

Why is the region is unavailable error reported for a SQL query after I use BR to restore the backup data?

If the cluster backed up using BR has TiFlash, TableInfo stores the TiFlash information when BR restores the backup data. If the cluster to be restored does not have TiFlash, the region is unavailable error is reported.

Does BR support in-place full recovery of some historical backup?

No. BR does not support in-place full recovery of some historical backup.

How can I use BR for incremental backup in the Kubernetes environment?

To get the commitTs field of the last BR backup, run the kubectl -n ${namespace} get bk ${name} command using kubectl. You can use the content of this field as --lastbackupts.

How can I convert BR backupTS to Unix time?

BR backupTS defaults to the latest timestamp obtained from PD before the backup starts. You can use pd-ctl tso timestamp to parse the timestamp to obtain an accurate value, or use backupTS >> 18 to quickly obtain an estimated value.

After BR restores the backup data, do I need to execute the ANALYZE statement on the table to update the statistics of TiDB on the tables and indexes?

BR does not back up statistics (except in v4.0.9). Therefore, after restoring the backup data, you need to manually execute ANALYZE TABLE or wait for TiDB to automatically execute ANALYZE.

In v4.0.9, BR backs up statistics by default, which consumes too much memory. To ensure that the backup process goes well, the backup for statistics is disabled by default starting from v4.0.10.

If you do not execute ANALYZE on the table, TiDB will fail to select the optimized execution plan due to inaccurate statistics. If query performance is not a key concern, you can ignore ANALYZE.

Can I use multiple BR processes at the same time to restore the data of a single cluster?

It is strongly not recommended to use multiple BR processes at the same time to restore the data of a single cluster for the following reasons:

  • When BR restores data, it modifies some global configurations of PD. Therefore, if you use multiple BR processes for data restore at the same time, these configurations might be mistakenly overwritten and cause abnormal cluster status.
  • BR consumes a lot of cluster resources to restore data, so in fact, running BR processes in parallel improves the restore speed only to a limited extent.
  • There has been no test for running multiple BR processes in parallel for data restore, so it is not guaranteed to succeed.

Was this page helpful?

Download PDFRequest docs changesAsk questions on Discord