如果选择不同的 pk 并按 pk 和限制排序，不要聚合整列

Question

我有一个 clickhouse table events 一年包含 5000 万行（可能重复）


    create table events (
        event LowCardinality(String),
        event_time DateTime64(3),
        uid       String
    ) engine=ReplacingMergeTree() toYYYYMM(event_time)
        primary key (event, event_time)
        order by (event, event_time, uid)

当我尝试 select 前 500 个不同的行时，Clickhouse 处理所有这些行 (50M)

    select distinct event, event_time 
    from events 
    where event='some_event' and event_time between '2020-02-24 00:00:00.000' AND '2021-02-24 00:00:00.000' 
    order by(event, event_time) desc limit 500

所以它基本上按事件分组，event_time 50M 行，然后再对其应用限制。是否可以强制clickhouse全程不读取数据？

Answer 1

CREATE TABLE events
(
    `event` LowCardinality(String),
    `event_time` DateTime,
    `uid` String
)
ENGINE = ReplacingMergeTree
PARTITION BY toYYYYMM(event_time)
ORDER BY (event, event_time, uid);


INSERT INTO events SELECT
    'ev',
    toDateTime('2020-01-01 00:00:00') + toIntervalSecond(number),
    randomPrintableASCII(5)
FROM numbers(30000000);
    
    
SELECT *
FROM
(
    SELECT event_time
    FROM events
    WHERE (event = 'ev') AND ((event_time >= '2020-01-01 00:00:00') AND (event_time <= '2021-01-01 00:00:00'))
    ORDER BY
        event DESC,
        event_time DESC
    LIMIT 1 BY event_time
)
LIMIT 500
.....

Elapsed: 0.008 sec. Processed 1.07 million rows

如果选择不同的 pk 并按 pk 和限制排序，不要聚合整列

Dont aggregate entire column if selecting distinct pk's with order by pk and limit

clickhouse