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
}
            
名称类型是否必选描述
table_namestring要切分的数据所在的表名。
split_sizeint64每个分片的近似大小,以百兆为单位。

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.
     */
     message SplitLocation {
         required string location = 1;
         required sint64 repeat = 2;
     }
     repeated SplitLocation locations = 4;
}
            
名称类型描述
consumedConsumedCapacity本次请求消耗的服务能力单元。
schemaPrimaryKeySchema该表的Schema,与建表时的Schema相同。
split_pointsrepeated 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。
locationsrepeated 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。

  • 本页导读 (1)
文档反馈