Postgres中按字符串标识符动态表分区

Question

我想按以前未知的值对 Postgres 中的表进行分区。在我的场景中，该值将是 device_id，它是一个字符串。

这是目前的情况：

表 'device_data' - 存储从设备发送的传感器数据，由 DDL 定义：

CREATE TABLE warehouse.device_data (
  id INTEGER PRIMARY KEY NOT NULL DEFAULT nextval('device_data_id_seq'::regclass),
  device_id TEXT NOT NULL,
  device_data BYTEA NOT NULL,
--   contains additional fields which are omitted for brevity
  received_at TIMESTAMP WITHOUT TIME ZONE DEFAULT now()
);

表目前拥有数百万条记录，查询需要花费大量时间。大多数查询都包含WHERE device_id='something'子句。

我想到的解决方案是为每个device_id.

Postgres 是否可以为每个创建表分区device_id？

我浏览了 Postgres 文档和我发现的几个示例，但它们都使用固定边界来创建分区。我的解决方案需要：

1. device_id首次遇到new 时动态创建新的表分区
2. 存储到现有分区，如果device_id已知并且该分区device_id已经存在

我希望使用表分区来完成这项工作，因为它允许跨多个device_ids进行查询。

Answer 1

I like the idea of dynamic partitioning. I do not know though how it will affect the performance as I have never used it.

Change the type of id to int default 0 and manually create the sequence to avoid multiple nextval() calls on a single insert:

create table device_data (
    id int primary key default 0,
    device_id text not null,
    device_data text not null, -- changed for tests
    received_at timestamp without time zone default now()
);
create sequence device_data_seq owned by device_data.id;

Use dynamic sql in the trigger function:

create or replace function before_insert_on_device_data()
returns trigger language plpgsql as $$
begin
    execute format(
        $f$
            create table if not exists %I (
            check (device_id = %L)
            ) inherits (device_data)
        $f$, 
        concat('device_data_', new.device_id), 
        new.device_id);
    execute format(
        $f$
            insert into %I
            values (nextval('device_data_seq'), %L, %L, default)
        $f$, 
        concat('device_data_', new.device_id), 
        new.device_id, 
        new.device_data);
    return null;
end $$;

create trigger before_insert_on_device_data
    before insert on device_data
    for each row execute procedure before_insert_on_device_data();

Test:

insert into device_data (device_id, device_data) values
    ('first', 'data 1'),
    ('second', 'data 1'),
    ('first', 'data 2'),
    ('second', 'data 2');

select * from device_data_first;

 id | device_id | device_data |        received_at         
----+-----------+-------------+----------------------------
  1 | first     | data 1      | 2016-10-18 19:50:40.179955
  3 | first     | data 2      | 2016-10-18 19:50:40.179955
(2 rows)

select * from device_data_second;

 id | device_id | device_data |        received_at         
----+-----------+-------------+----------------------------
  2 | second    | data 1      | 2016-10-18 19:50:40.179955
  4 | second    | data 2      | 2016-10-18 19:50:40.179955
(2 rows)