記錄一次ClickHouse的踩坑經歷

最近在做Metabase和ClickHouse集成，使用社區開發的數據驅動把ClickHouse集成到了Metabase。 大部分查詢是沒有問題，只時間維度的聚合查詢會報以下異常：

DB::Exception: Column fdate is not under aggregate function and not in GROUP BY.

詳細使用場景如下：

一、ClickHouse的版本

Connected to ClickHouse server version 20.1.4 revision 54431.

二、查詢表結構

2.1 本地表創建sql

create table if not exists dw_live.dw_player_dim_hour_local on cluster test_cluster_two_replicas (
fdate Datetime,
num Int64 default 0
)
ENGINE = ReplicatedMergeTree('/clickhouse/tables/{cluster}-{shard}/dw_live-dw_player_dim_hour_local', '{replica}')
PARTITION BY (fdate) 
PRIMARY KEY (fdate) 
ORDER BY (fdate) 
TTL fdate + toIntervalDay(120) 
SETTINGS index_granularity = 8192, storage_policy = 'round', merge_with_ttl_timeout = 3600

2.2 分布式表創建sql

create table if not exists dw_live.dw_player_dim_hour(
fdate Datetime,
num Int64 default 0
)
ENGINE = Distributed(test_cluster_two_replicas, dw_live, dw_player_dim_hour_local, rand())

三、發生異常的使用場景

在Metabase頁面點擊dwlive.dwplayerdimhour表的時間時間維度聚合查詢，監控后台異常日志，頁面向ClickHouse請求的查詢SQL是(查詢的是分布式表)：

SELECT 
    toStartOfHour(toDateTime(`dw_live`.`dw_player_dim_hour`.`fdate`)) AS `fdate`, 
    count() AS `count` 
FROM `dw_live`.`dw_player_dim_hour` 
GROUP BY toStartOfHour(toDateTime(`dw_live`.`dw_player_dim_hour`.`fdate`)) 
ORDER BY toStartOfHour(toDateTime(`dw_live`.`dw_player_dim_hour`.`fdate`)) ASC

將該SQL粘貼到ClickHouse的命令行窗口執行，同樣報以下異常：

DB::Exception: Column fdate is not under aggregate function and not in GROUP BY.

同樣的查詢語句，查詢本地表返回結果正常。查詢本地表的SQL是：

SELECT 
    toStartOfHour(toDateTime(`dw_live`.`dw_player_dim_hour_local`.`fdate`)) AS `fdate`, 
    count() AS `count` 
FROM `dw_live`.`dw_player_dim_hour` 
GROUP BY toStartOfHour(toDateTime(`dw_live`.`dw_player_dim_hour_local`.`fdate`)) 
ORDER BY toStartOfHour(toDateTime(`dw_live`.`dw_player_dim_hour_local`.`fdate`)) ASC

四、結論和解決方法

到此判定是ClickHouse的Bug，升級到20.3.18，沒有再復現該查詢異常。