Doc Menu

Upgrade TiDB Using TiUP

This document is targeted for users who want to upgrade from TiDB 3.0 or 3.1 versions to TiDB 4.0 versions, or from TiDB 4.0 to a later version.

If you have deployed the TiDB cluster using TiDB Ansible, you can use TiUP to import the TiDB Ansible configuration and perform the upgrade.

Upgrade caveat

  • After the upgrade, rolling back to 3.0 or earlier versions is not supported.
  • To update versions earlier than 3.0 to 4.0, first update this version to 3.0 using TiDB Ansible, and use TiUP to import the TiDB Ansible configuration and update the 3.0 version to 4.0.
  • After the TiDB Ansible configuration is imported into and managed by TiUP, you can no longer operate on the cluster using TiDB Ansible. Otherwise, conflicts might occur because of the inconsistent metadata.
  • Currently, you cannot import the TiDB Ansible configuration if the cluster deployed using TiDB Ansible meets one of the following situations:
    • The TLS encryption is enabled for the cluster.
    • This is a pure key-value cluster (cluster with no TiDB instance).
    • Kafka is enabled for the cluster.
    • Spark is enabled for the cluster.
    • Lightning / Importer is enabled for the cluster.
    • You still use the 'push' method to collect monitoring metrics (since v3.0, pull is the default mode, which is supported if you have not modified this mode).
    • In the inventory.ini configuration file, the node_exporter or blackbox_exporter item of the machine is set to non-default ports through node_exporter_port or blackbox_exporter_port, which is compatible if you have unified the configuration in the group_vars directory.
  • Support upgrading the versions of TiDB Binlog, TiCDC, TiFlash, and other components.
  • Before you upgrade from v2.0.6 or earlier to v4.0.0 or later, you must make sure that no DDL operations are running in the cluster, especially the Add Index operation that is time-consuming. Perform the upgrade after all DDL operations are completed.
  • Starting from v2.1, TiDB enables parallel DDL. Therefore, clusters older than v2.0.1 cannot be upgraded to v4.0.0 or later via a direct rolling upgrade. Instead, you can choose one of the following solutions:
    • Upgrade directly from TiDB v2.0.1 or earlier to v4.0.0 or later in planned downtime
    • Perform a rolling upgrade from the current version to v2.0.1 or a later 2.0 version, then perform another rolling upgrade to v4.0.0 or later

Note:

Do not execute any DDL request during the upgrade, otherwise an undefined behavior issue might occur.

Install TiUP on the control machine

  1. Execute the following command on the control machine to install TiUP:

    curl --proto '=https' --tlsv1.2 -sSf https://tiup-mirrors.pingcap.com/install.sh | sh
  2. Redeclare the global environment variables:

    source .bash_profile
  3. Check whether TiUP is installed:

    which tiup
  4. Install the TiUP cluster tool:

    tiup cluster

If you have installed TiUP before, execute the following command to update TiUP to the latest version:

Note:

If the result of tiup --version shows that your TiUP version is earlier than v1.0.0, run tiup update --self first to update the TiUP version before running the following command.

tiup update cluster

Import TiDB Ansible and the inventory.ini configuration to TiUP

Note:

  • Currently, TiUP only supports the systemd supervision method of a process. If you have previously selected the supervise method when deploying TiDB with TiDB Ansible, you need to modify the supervision method from supervise to systemd according to Deploy TiDB Using TiDB Ansible.
  • If the original cluster is deployed using TiUP, you can skip this step.
  • Currently, the inventory.ini configuration file is identified by default. If your configuration file uses another name, specify this name.
  • Ensure that the state of the current cluster is consistent with the topology in inventory.ini; that components of the cluster are operating normally. Otherwise, the cluster metadata becomes abnormal after the import.
  • If multiple different inventory.ini files and TiDB clusters are managed in one TiDB Ansible directory, when importing one of the clusters into TiUP, you need to specify --no-backup to avoid moving the Ansible directory to the TiUP management directory.

Import the TiDB Ansible cluster to TiUP

  1. Execute the following command to import the TiDB Ansible cluster into TiUP (for example, in the /home/tidb/tidb-ansible path).

    tiup cluster import -d /home/tidb/tidb-ansible
  2. After executing the above import command, if the Inventory information of the cluster is parsed successfully, the following prompt appears:

    tiup cluster import -d /home/tidb/tidb-ansible/
    Found inventory file /home/tidb/tidb-ansible/inventory.ini, parsing...
    Found cluster "ansible-cluster" (v3.0.12), deployed with user tidb.
    Prepared to import TiDB v3.0.12 cluster ansible-cluster.
    Do you want to continue? [y/N]:
  3. After checking that the parsed cluster name and the version are correct, enter y to continue the import process.

    • If an error occurs when parsing the Inventory information, the import process is stopped, which does not have any impact on the original Ansible deployment method. Then you need to adjust and retry the process according to the error prompt.

    • If the original cluster name in Ansible is the same with any existing cluster name in TiUP, a warning message is returned with a new cluster name. Therefore, do not repeatedly import the same cluster, which results in multiple names for the same cluster in TiUP.

After the import is complete, you can check the current cluster status by executing the tiup cluster display <cluster-name> command to verify the import result. Because the display command is used to query the real-time status of each node, it might take a little time to execute the command.

Edit TiUP topology configuration file

Note:

You can skip this step for the following situations:

  • The configuration parameters in the original cluster have not been modified.
  • You want to use the default parameters of 4.0 after the upgrade.
  1. Enter ~/.tiup/storage/cluster/clusters/{cluster_name}/config, the backup directory of TiDB Ansible, and confirm the modified parameters in the configuration template.

  2. Enter the vi editing mode of the topology file:

    tiup cluster edit-config <cluster-name>
  3. See the configuration template format of topology and fill in the modified parameters of the original cluster in the server_configs section of the topology file.

    Even if the label has been configured for the cluster, you also need to fill in the label in the configuration according to the format in the template. In later versions, the label will be automatically imported.

    After the modification is completed, execute the wq command to save the change and exit the editing mode. Enter Y to confirm the change.

Note:

Before upgrading to v4.0, confirm that the parameters modified in v3.0 are compatible in v4.0. See configuration template for details.

If the TiUP version <= v1.0.8, TiUP might not correctly obtain the data directory of TiFlash, and you need to check whether data_dir and path configured in TiFlash is consistent. If not, configure data_dir of TiFlash to the same value as path by taking the following steps:

  1. Execute tiup cluster edit-config <cluster-name> to modify the configuration file.

  2. Modify the corresponding data_dir value of TiFlash:

      tiflash_servers:
            - host: 10.0.1.14
              data_dir: data/tiflash-11315 # Modify it to the `path` value of the TiFlash configuration file

Perform a rolling upgrade to the TiDB cluster

This section describes how to perform a rolling upgrade to the TiDB cluster and how to verify the version after the upgrade.

Upgrade the TiDB cluster to a specified version

tiup cluster upgrade <cluster-name> <version>

For example, if you want to upgrade the cluster to v4.0.0:

tiup cluster upgrade <cluster-name> v4.0.0

Performing a rolling upgrade to the cluster will upgrade all components one by one. During the upgrade of TiKV, all leaders in a TiKV instance are evicted before stopping the instance. The default timeout time is 5 minutes. The instance is directly stopped after this timeout time.

To perform the upgrade immediately without evicting the leader, specify --force in the command above. This method causes performance jitter but not data loss.

To keep a stable performance, make sure that all leaders in a TiKV instance are evicted before stopping the instance. You can set --transfer-timeout to a super large value, for example, --transfer-timeout 100000000 (unit: second).

Verify the cluster version

Execute the display command to view the latest cluster version TiDB Version:

tiup cluster display <cluster-name>
Starting /home/tidblk/.tiup/components/cluster/v1.0.0/cluster display <cluster-name>
TiDB Cluster: <cluster-name>
TiDB Version: v4.0.0

Note:

By default, TiUP and TiDB (starting from v4.0.2) share usage details with PingCAP to help understand how to improve the product. For details about what is shared and how to disable the sharing, see Telemetry.

FAQ

This section describes common problems encountered when updating the TiDB cluster using TiUP.

If an error occurs and the upgrade is interrupted, how to resume the upgrade after fixing this error?

Re-execute the tiup cluster upgrade command to resume the upgrade. The upgrade operation restarts the nodes that have been previously upgraded. In subsequent 4.0 versions, TiDB will support resuming the upgrade from the interrupted point.

The evict leader has waited too long during the upgrade. How to skip this step for a quick upgrade?

You can specify --force. Then the processes of transferring PD leader and evicting TiKV leader are skipped during the upgrade. The cluster is directly restarted to update the version, which has a great impact on the cluster that runs online. Here is the command:

tiup cluster upgrade <cluster-name> v4.0.0 --force

How to update the version of tools such as pd-ctl after upgrading the TiDB cluster?

Currently, TiUP does not update and manage the version of tools. If you need the tool package of the latest version, directly download the TiDB package and replace {version} with the corresponding version such as v4.0.0. Here is the download address:

https://download.pingcap.org/tidb-{version}-linux-amd64.tar.gz

Failure to upgrade the TiFlash component during the cluster upgrade

Before v4.0.0-rc.2, TiFlash might have some incompatibility issues. This might cause problems when you upgrade a cluster that includes the TiFlash component to v4.0.0-rc.2 or a later version. If so, contact the R&D support.

TiDB 4.0 compatibility changes

  • If you set the value of the oom-action parameter to cancel, when the query statement triggers the OOM threshold, the statement is killed. In v4.0, in addition to select, DML statements such as insert/update/delete might also be killed.
  • TiDB v4.0 supports the length check for table names. The length limit is 64 characters. If you rename a table after the upgrade and the new name exceeds this length limit, an error is reported. v3.0 and earlier versions do not have this error reporting.
  • TiDB v4.0 supports the length check for partition names of the partitioned tables. The length limit is 64 characters. After the upgrade, if you create or alter a partitioned table with a partition name that exceeds the length limit, an error is expected to occur in 4.0 versions, but not in 3.0 and earlier versions.
  • In v4.0, the format of the explain execution plan is improved. Pay attention to any automatic analysis program that is customized for explain.
  • TiDB v4.0 supports Read Committed isolation level. After upgrading to v4.0, setting the isolation level to READ-COMMITTED in a pessimistic transaction takes effect. In 3.0 and earlier versions, the setting does not take effect.
  • In v4.0, executing alter reorganize partition returns an error. In earlier versions, no error is reported because only the syntax is supported and the statement is not taking any effect.
  • In v4.0, creating linear hash partition or subpartition tables does not take effect and they are converted to regular tables. In earlier versions, they are converted to regular partitioned tables.