Placement Rules in SQL

Placement Rules in SQL is a feature that enables you to specify where data is stored in a TiKV cluster using SQL interfaces. Using this feature, tables and partitions are scheduled to specific regions, data centers, racks, or hosts. This is useful for scenarios including optimizing a high availability strategy with lower cost, ensuring that local replicas of data are available for local stale reads, and adhering to data locality requirements.

The detailed user scenarios are as follows:

• Merge multiple databases of different applications to reduce the cost on database maintenance
• Increase replica count for important data to improve the application availability and data reliability
• Store new data into NVMe storage and store old data into SSDs to lower the cost on data archiving and storage
• Schedule the leaders of hotspot data to high-performance TiKV instances
• Separate cold data to lower-cost storage mediums to improve cost efficiency

Specify placement rules

To specify placement rules, first create a placement policy using CREATE PLACEMENT POLICY:

CREATE PLACEMENT POLICY myplacementpolicy PRIMARY_REGION="us-east-1" REGIONS="us-east-1,us-west-1";

Then attach the policy to a table or partition using either CREATE TABLE or ALTER TABLE. Then, the placement rules are specified on the table or the partition:

CREATE TABLE t1 (a INT) PLACEMENT POLICY=myplacementpolicy;
CREATE TABLE t2 (a INT);
ALTER TABLE t2 PLACEMENT POLICY=myplacementpolicy;

A placement policy is not associated with any database schema and has the global scope. Therefore, assigning a placement policy does not require any additional privileges over the CREATE TABLE privilege.

To modify a placement policy, you can use ALTER PLACEMENT POLICY, and the changes will propagate to all objects assigned with the corresponding policy.

ALTER PLACEMENT POLICY myplacementpolicy FOLLOWERS=5;

To drop policies that are not attached to any table or partition, you can use DROP PLACEMENT POLICY:

DROP PLACEMENT POLICY myplacementpolicy;

View current placement rules

If a table has placement rules attached, you can view the placement rules in the output of SHOW CREATE TABLE. To view the definition of the policy available, execute SHOW CREATE PLACEMENT POLICY:

tidb> SHOW CREATE TABLE t1\G
*************************** 1. row ***************************
       Table: t1
Create Table: CREATE TABLE `t1` (
  `a` int(11) DEFAULT NULL
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin /*T![placement] PLACEMENT POLICY=`myplacementpolicy` */
1 row in set (0.00 sec)

tidb> SHOW CREATE PLACEMENT POLICY myplacementpolicy\G
*************************** 1. row ***************************
       Policy: myplacementpolicy
Create Policy: CREATE PLACEMENT POLICY myplacementpolicy PRIMARY_REGION="us-east-1" REGIONS="us-east-1,us-west-1"
1 row in set (0.00 sec)

You can also view definitions of placement policies using the INFORMATION_SCHEMA.PLACEMENT_POLICIES table.

tidb> select * from information_schema.placement_policies\G
***************************[ 1. row ]***************************
POLICY_ID            | 1
CATALOG_NAME         | def
POLICY_NAME          | p1
PRIMARY_REGION       | us-east-1
REGIONS              | us-east-1,us-west-1
CONSTRAINTS          | 
LEADER_CONSTRAINTS   | 
FOLLOWER_CONSTRAINTS | 
LEARNER_CONSTRAINTS  | 
SCHEDULE             | 
FOLLOWERS            | 4
LEARNERS             | 0
1 row in set

The information_schema.tables and information_schema.partitions tables also include a column for tidb_placement_policy_name, which shows all objects with placement rules attached:

SELECT * FROM information_schema.tables WHERE tidb_placement_policy_name IS NOT NULL;
SELECT * FROM information_schema.partitions WHERE tidb_placement_policy_name IS NOT NULL;

Rules that are attached to objects are applied asynchronously. To view the current scheduling progress of placement, use SHOW PLACEMENT.

Option reference

In addition to the placement options above, you can also use the advance configurations. For details, see Advance placement options.

Examples

Increase the number of replicas

The default configuration of max-replicas is 3. To increase this for a specific set of tables, you can use a placement policy as follows:

CREATE PLACEMENT POLICY fivereplicas FOLLOWERS=4;
CREATE TABLE t1 (a INT) PLACEMENT POLICY=fivereplicas;

Note that the PD configuration includes the leader and follower count, thus 4 followers + 1 leader equals 5 replicas in total.

To expand on this example, you can also use PRIMARY_REGION and REGIONS placement options to describe the placement for the followers:

CREATE PLACEMENT POLICY eastandwest PRIMARY_REGION="us-east-1" REGIONS="us-east-1,us-east-2,us-west-1" SCHEDULE="MAJORITY_IN_PRIMARY" FOLLOWERS=4;
CREATE TABLE t1 (a INT) PLACEMENT POLICY=eastandwest;

The SCHEDULE option instructs TiDB on how to balance the followers. The default schedule of EVEN ensures a balance of followers in all regions.

To ensure that enough followers are placed in the primary region (us-east-1) so that quorum can be achieved, you can use the MAJORITY_IN_PRIMARY schedule. This schedule helps provide lower latency transactions at the expense of some availability. If the primary region fails, MAJORITY_IN_PRIMARY cannot provide automatic failover.

Assign placement to a partitioned table

In addition to assigning placement options to tables, you can also assign the options to table partitions. For example:

CREATE PLACEMENT POLICY p1 FOLLOWERS=5;
CREATE PLACEMENT POLICY europe PRIMARY_REGION="eu-central-1" REGIONS="eu-central-1,eu-west-1";
CREATE PLACEMENT POLICY northamerica PRIMARY_REGION="us-east-1" REGIONS="us-east-1";

SET tidb_enable_list_partition = 1;
CREATE TABLE t1 (
  country VARCHAR(10) NOT NULL,
  userdata VARCHAR(100) NOT NULL
) PLACEMENT POLICY=p1 PARTITION BY LIST COLUMNS (country) (
  PARTITION pEurope VALUES IN ('DE', 'FR', 'GB') PLACEMENT POLICY=europe,
  PARTITION pNorthAmerica VALUES IN ('US', 'CA', 'MX') PLACEMENT POLICY=northamerica,
  PARTITION pAsia VALUES IN ('CN', 'KR', 'JP')
);

If a partition has no attached policies, it tries to apply possibly existing policies on the table. For example, the pEurope partition will apply the europe policy, but the pAsia partition will apply the p1 policy from table t1. If t1 has no assigned policies, pAsia will not apply any policy, too.

You can also alter the placement policies assigned to a specific partition. For example:

ALTER TABLE t1 PARTITION pEurope PLACEMENT POLICY=p1;

Set the default placement for a schema

You can directly attach the default placement rules to a database schema. This works similar to setting the default character set or collation for a schema. Your specified placement options apply when no other options are specified. For example:

CREATE PLACEMENT POLICY p1 PRIMARY_REGION="us-east-1" REGIONS="us-east-1,us-east-2";  -- Create placement policies

CREATE PLACEMENT POLICY p2 FOLLOWERS=4;

CREATE PLACEMENT POLICY p3 FOLLOWERS=2;

CREATE TABLE t1 (a INT);  -- Creates a table t1 with no placement options.

ALTER DATABASE test PLACEMENT POLICY=p2;  -- Changes the default placement option, and does not apply to the existing table t1.

CREATE TABLE t2 (a INT);  -- Creates a table t2 with the default placement policy p2.

CREATE TABLE t3 (a INT) PLACEMENT POLICY=p1;  -- Creates a table t3 without the default policy p2, because this statement has specified another placement rule.

ALTER DATABASE test PLACEMENT POLICY=p3;  -- Changes the default policy, and does not apply to existing tables.

CREATE TABLE t4 (a INT);  -- Creates a table t4 with the default policy p3.

ALTER PLACEMENT POLICY p3 FOLLOWERS=3; -- The table with policy p3 (t4) will have FOLLOWERS=3.

Note that this is different from the inheritance between partitions and tables, where changing the policy of tables will affect their partitions. Tables inherit the policy of schema only when they are created without policies attached, and modifying the policies of schemas does not affect created tables.

Advanced placement options

The placement options PRIMARY_REGION, REGIONS, and SCHEDULE meet the basic needs of data placement at the loss of some flexibility. For more complex scenarios with the need for higher flexibility, you can also use the advanced placement options of CONSTRAINTS and FOLLOWER_CONSTRAINTS. You cannot specify the PRIMARY_REGION, REGIONS, or SCHEDULE option with the CONSTRAINTS option at the same time. If you specify both at the same time, an error will be returned.

For example, to set constraints that data must reside on a TiKV store where the label disk must match a value:

CREATE PLACEMENT POLICY storageonnvme CONSTRAINTS="[+disk=nvme]";
CREATE PLACEMENT POLICY storageonssd CONSTRAINTS="[+disk=ssd]";
CREATE PLACEMENT POLICY companystandardpolicy CONSTRAINTS="";

CREATE TABLE t1 (id INT, name VARCHAR(50), purchased DATE)
PLACEMENT POLICY=companystandardpolicy
PARTITION BY RANGE( YEAR(purchased) ) (
  PARTITION p0 VALUES LESS THAN (2000) PLACEMENT POLICY=storageonssd,
  PARTITION p1 VALUES LESS THAN (2005),
  PARTITION p2 VALUES LESS THAN (2010),
  PARTITION p3 VALUES LESS THAN (2015),
  PARTITION p4 VALUES LESS THAN MAXVALUE PLACEMENT POLICY=storageonnvme
);

You can either specify constraints in list format ([+disk=ssd]) or in dictionary format ({+disk=ssd: 1,+disk=nvme: 2}).

In list format, constraints are specified as a list of key-value pairs. The key starts with either a + or a -. +disk=ssd indicates that the label disk must be set to ssd, and -disk=nvme indicates that the label disk must not be nvme.

In dictionary format, constraints also indicate a number of instances that apply to that rule. For example, FOLLOWER_CONSTRAINTS="{+region=us-east-1: 1,+region=us-east-2: 1,+region=us-west-1: 1}"; indicates that 1 follower is in us-east-1, 1 follower is in us-east-2 and 1 follower is in us-west-1. For another example, FOLLOWER_CONSTRAINTS='{"+region=us-east-1,+disk=nvme":1,"+region=us-west-1":1}'; indicates that 1 follower is in us-east-1 with an nvme disk, and 1 follower is in us-west-1.

Compatibility with tools

Known limitations

The following known limitations are as follows:

• Temporary tables do not support placement options.
• Syntactic sugar rules are permitted for setting PRIMARY_REGION and REGIONS. In the future, we plan to add varieties for PRIMARY_RACK, PRIMARY_ZONE, and PRIMARY_HOST. See issue #18030.
• TiFlash learners are not configurable through Placement Rules syntax.
• Placement rules only ensure that data at rest resides on the correct TiKV store. The rules do not guarantee that data in transit (via either user queries or internal operations) only occurs in a specific region.