文档

ComputeSplitPointsBySize

更新时间:

调用ComputeSplitPointsBySize接口将全表的数据在逻辑上划分成接近指定大小的若干分片,返回这些分片之间的分割点以及分片所在机器的提示。一般用于计算引擎规划并发度等执行计划。

请求消息结构

message ComputeSplitPointsBySizeRequest {
    required string table_name = 1;
    required int64 split_size = 2; // in 100MB
    optional int64 split_size_unit_in_byte = 3;
    optional int32 split_point_limit = 4;
}

名称

类型

是否必选

描述

table_name

string

要切分的数据所在的表名。

split_size

int64

每个分片的近似大小,以百兆为单位。

split_size_unit_in_byte

int64

指定分割大小的单位,以便在分割点计算时使用正确的单位,并确保计算结果的准确性。

split_point_limit

int32

指定对分割点数量的限制,以便在进行分割点计算时控制返回的分割点数量。

响应消息结构

message ComputeSplitPointsBySizeResponse {
    required ConsumedCapacity consumed = 1;
    repeated PrimaryKeySchema schema = 2;

    /**
     * Split points between splits, in the increasing order
     *
     * A split is a consecutive range of primary keys,
     * whose data size is about split_size specified in the request.
     * The size could be hard to be precise.
     *
     * A split point is an array of primary-key column w.r.t. table schema,
     * which is never longer than that of table schema.
     * Tailing -inf will be omitted to reduce transmission payloads.
     */
    repeated bytes split_points = 3;

    /**
     * Locations where splits lies in.
     *
     * By the managed nature of TableStore, these locations are no more than hints.
     * If a location is not suitable to be seen, an empty string will be placed.
     */
     repeated SplitLocation locations = 4;
}

名称

类型

描述

consumed

ConsumedCapacity

本次请求消耗的服务能力单元。

schema

PrimaryKeySchema

该表的Schema,与建表时的Schema相同。

split_points

repeated bytes

分片之间的分割点。分割点之间保证单调增。每个分割点都是以Plainbuffer编码的行,并且只有primary-key项。为了减少传输的数据量,分割点最后的-inf不会传输。

locations

repeated SplitLocation

分割点所在机器的提示。可以为空。

例如有一张表有三列主键,其中首列主键类型为string。调用该API后得到5个分片,分别为(-inf,-inf,-inf)("a",-inf,-inf)("a",-inf,-inf)("b",-inf,-inf)("b",-inf,-inf)("c",-inf,-inf)("c",-inf,-inf)("d",-inf,-inf)("d",-inf,-inf)(+inf,+inf,+inf)。前三个落在"machine-A",后两个落在"machine-B"。那么,split_points为(示意)[("a"),("b"),("c"),("d")],而locations为(示意)"machine-A"*3, "machine-B"*2

使用SDK

Java SDK:指定大小计算分片

服务能力单元消耗

消耗的读服务能力单元数量与分片的数量相同。不消耗写服务能力单元。