Sign InTry Free

Best Practices for Read-Only Storage Nodes

This document introduces how to configure read-only storage nodes and how to direct backup, analysis, testing, and other traffic to these nodes. In this way, loads with high tolerance for delay can be physically isolated from important online services.


1. Specify some TiKV nodes as read-only

To specify some TiKV nodes as read-only, you can mark these nodes with a special label (use $ as the prefix of the label key). Unless you explicitly specify these nodes to store some data using Placement Rules, PD does not schedule any data to these nodes.

You can configure a read-only node by running the tiup cluster edit-config command:

tikv_servers: - host: ... ... labels: $mode: readonly

2. Use Placement Rules to store data on read-only nodes as learners

  1. Run the pd-ctl config placement-rules command to export the default Placement Rules:

    pd-ctl config placement-rules rule-bundle load --out="rules.json"

    If you have not configured Placement Rules before, the output is as follows:

    [ { "group_id": "pd", "group_index": 0, "group_override": false, "rules": [ { "group_id": "pd", "id": "default", "start_key": "", "end_key": "", "role": "voter", "count": 3 } ] } ]
  2. Store all data on the read-only nodes as a learner. The following example is based on the default configuration:

    [ { "group_id": "pd", "group_index": 0, "group_override": false, "rules": [ { "group_id": "pd", "id": "default", "start_key": "", "end_key": "", "role": "voter", "count": 3 }, { "group_id": "pd", "id": "readonly", "start_key": "", "end_key": "", "role": "learner", "count": 1, "label_constraints": [ { "key": "$mode", "op": "in", "values": [ "readonly" ] } ], "version": 1 } ] } ]
  3. Use the pd-ctl config placement-rules command to write the preceding configurations to PD:

    pd-ctl config placement-rules rule-bundle save --in="rules.json"

3. Use Follower Read to read data from read-only nodes

3.1 Use Follower Read in TiDB

To read data from read-only nodes when using TiDB, you can set the system variable tidb_replica_read to learner:

set tidb_replica_read=learner;

3.2 Use Follower Read in TiSpark

To read data from read-only nodes when using TiSpark, you can set the configuration item spark.tispark.replica_read to learner in the Spark configuration file:

spark.tispark.replica_read learner

3.3 Use Follower Read when backing up cluster data

To read data from read-only nodes when backing up cluster data, you can specify the --replica-read-label option in the br command line. Note that when running the following command in shell, you need to use single quotes to wrap the label to prevent $ from being parsed.

br backup full ... --replica-read-label '$mode:readonly'

Was this page helpful?

Download PDFRequest docs changesAsk questions on DiscordEdit this page
One-stop & interactive experience of TiDB's capabilities WITHOUT registration.
TiDB Dedicated
TiDB Serverless
Get Demo
Get Started
© 2024 PingCAP. All Rights Reserved.
Privacy Policy.