We support three types of hints in DBSP today. All three of them apply to an individual stream (an input to a join operator):

    /// Apply the specified policy to the stream instead of selecting one automatically.
    /// `None` clears any previously set hint.
    Policy(Option<PartitioningPolicy>),

    /// Choose a policy for the stream as if its total size, i.e., the number of records
    /// in the stream, is equal to the specified value.
    /// Setting size hint to `None` tells the balancer to use the actual size of the stream
    /// measured at runtime.
    Size(Option<usize>),

    /// Choose a policy for the stream as if its skew, i.e., the ratio of the largest
    /// to the average partition size, is equal to the specified value.
    /// Setting skew hint to `None` tells the balancer to use the actual skew of the stream
    /// measured at runtime.
    Skew(Option<F64>),

So something like:

SELECT
...
FROM
  tableName /*+ broadcast */
JOIN
  tableName

SELECT
...
FROM
  tableName /*+ shard */
JOIN
  tableName

SELECT
...
FROM
  tableName /*+ balance */
JOIN
  tableName

SELECT
...
FROM
  tableName /*+ size(5) */
JOIN
  tableName /*+ size(1000000) */

SELECT
...
FROM
  tableName /*+ size(5) */
JOIN
  tableName /* + size(1000000), skew(10.0) */

[SQL] Add support for optimizer hints #6160

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions