Maintain a TiFlash Cluster

This document describes how to perform common operations when you maintain a TiFlash cluster, including checking the TiFlash version, taking TiFlash nodes down, and troubleshooting TiFlash. This document also introduces critical logs and a system table of TiFlash.

Check the TiFlash version

There are two ways to check the TiFlash version:

• If the binary file name of TiFlash is tiflash, you can check the version by executing the ./tiflash version command.

  However, to execute the above command, you need to add the directory path which includes the libtiflash_proxy.so dynamic library to the LD_LIBRARY_PATH environment variable. This is because the running of TiFlash relies on the libtiflash_proxy.so dynamic library.

  For example, when tiflash and libtiflash_proxy.so are in the same directory, you can first switch to this directory, and then use the following command to check the TiFlash version:

  Copy

  LD_LIBRARY_PATH=./ ./tiflash version

• Check the TiFlash version by referring to the TiFlash log. For the log path, see the [logger] part in the tiflash.toml file. For example:

  <information>: TiFlash version: TiFlash 0.2.0 master-375035282451103999f3863c691e2fc2

Take a TiFlash node down

Taking a TiFlash node down differs from Scaling in a TiFlash node in that the former doesn't remove the node in TiDB Ansible; instead, it just safely shuts down the TiFlash process.

Follow the steps below to take a TiFlash node down:

Note:

After you take the TiFlash node down, if the number of the remaining nodes in the TiFlash cluster is greater than or equal to the maximum replicas of all data tables, you can go directly to step 3.

1. If the number of replicas of tables is greater than or equal to that of the remaining TiFlash nodes in the cluster, execute the following command on these tables in the TiDB client:

   Copy

   alter table <db-name>.<table-name> set tiflash replica 0;

2. To ensure that the TiFlash replicas of these tables are removed, see Check the Replication Progress. If you cannot view the replication progress of the related tables, it means that the replicas are removed.

3. Input the store command into pd-ctl (the binary file is in resources/bin of the tidb-ansible directory) to view the store id of the TiFlash node.

4. Input store delete <store_id> into pd-ctl. Here <store_id> refers to the store id in step 3.

5. When the corresponding store of the node disappears, or when state_name is changed to Tombstone, stop the TiFlash process.

Note:

If you don't cancel all tables replicated to TiFlash before all TiFlash nodes stop running, you need to manually delete the replication rules in PD. Or you cannot successfully take the TiFlash node down.

To manually delete the replication rules in PD, take the following steps:

1. Query all the data replication rules related to TiFlash in the current PD instance:

   Copy

   curl http://<pd_ip>:<pd_port>/pd/api/v1/config/rules/group/tiflash

   [
     {
       "group_id": "tiflash",
       "id": "table-45-r",
       "override": true,
       "start_key": "7480000000000000FF2D5F720000000000FA",
       "end_key": "7480000000000000FF2E00000000000000F8",
       "role": "learner",
       "count": 1,
       "label_constraints": [
         {
           "key": "engine",
           "op": "in",
           "values": [
             "tiflash"
           ]
         }
       ]
     }
   ]

2. Delete all the data replication rules related to TiFlash. The following example command deletes the rule whose id is table-45-r:

   Copy

   curl -v -X DELETE http://<pd_ip>:<pd_port>/pd/api/v1/config/rule/tiflash/table-45-r

TiFlash troubleshooting

This section describes some commonly encountered issues when using TiFlash, the reasons, and the solutions.

TiFlash replica is always unavailable

This is because TiFlash is in an abnormal state caused by configuration errors or environment issues. Take the following steps to identify the faulty component:

1. Check whether PD enables the Placement Rules feature (to enable the feature, see the step 2 of Add TiFlash component to an existing TiDB cluster:

   Copy

   echo 'config show replication' | /path/to/pd-ctl -u http://<pd-ip>:<pd-port>

   The expected result is "enable-placement-rules": "true".

2. Check whether the TiFlash process is working correctly by viewing UpTime on the TiFlash-Summary monitoring panel.

3. Check whether the TiFlash proxy status is normal through pd-ctl.

   Copy

   echo "store" | /path/to/pd-ctl -u http://<pd-ip>:<pd-port>

   The TiFlash proxy's store.labels includes information such as {"key": "engine", "value": "tiflash"}. You can check this information to confirm a TiFlash proxy.

4. Check whether pd buddy can correctly print the logs (the log path is the value of log in the [flash.flash_cluster] configuration item; the default log path is under the tmp directory configured in the TiFlash configuration file).

5. Check whether the value of max-replicas in PD is less than or equal to the number of TiKV nodes in the cluster. If not, PD cannot replicate data to TiFlash:

   Copy

   echo 'config show replication' | /path/to/pd-ctl -u http://<pd-ip>:<pd-port>

   Reconfirm the value of max-replicas.

6. Check whether the remaining disk space of the machine (where store of the TiFlash node is) is sufficient. By default, when the remaining disk space is less than 20% of the store capacity (which is controlled by the low-space-ratio parameter), PD cannot schedule data to this TiFlash node.

TiFlash query time is unstable, and the error log prints many Lock Exception messages

This is because large amounts of data are written to the cluster, which causes that the TiFlash query encounters a lock and requires query retry.

You can set the query timestamp to one second earlier in TiDB. For example, if the current time is '2020-04-08 20:15:01', you can execute set @@tidb_snapshot='2020-04-08 20:15:00'; before you execute the query. This makes less TiFlash queries encounter a lock and mitigates the risk of unstable query time.

Some queries return the Region Unavailable error

If the load pressure on TiFlash is too heavy and it causes that TiFlash data replication falls behind, some queries might return the Region Unavailable error.

In this case, you can balance the load pressure by adding more TiFlash nodes.

Data file corruption

Take the following steps to handle the data file corruption:

1. Refer to Take a TiFlash node down to take the corresponding TiFlash node down.
2. Delete the related data of the TiFlash node.
3. Redeploy the TiFlash node in the cluster.

TiFlash critical logs

You can find the beginning or the end of a Coprocessor request, and then locate the related logs of the Coprocessor request through the thread ID printed at the start of the log.

TiFlash system table

The column names and their descriptions of the information_schema.tiflash_replica system table are as follows: