使用 sql 查询对 clickhouse 中的表进行非规范化

Question

源表的格式

CREATE TABLE IF NOT EXISTS src_table (
    str_1 String,
    str_2 String,
    metric_1 UInt64,
    metric_2 UInt8
) ENGINE = Log

要反规范化的列是 str_2 和非规范化表

CREATE TABLE IF NOT EXISTS denorm_table (
    dt Date,
    str_1 String,
    attr_1 UInt64,
    attr_2 UInt64,
    ......
    attr_1000 UInt64,
    attr_1001 UInt8,
    attr_1002 UInt8,
    .....
    attr_2000 UInt8
) ENGINE = MergeTree PARTITION BY (dt) ORDER BY (dt, str_1) SETTINGS index_granularity=8192

假设 str_2 列有 1000 个不同的值 (1 ... 1000)，并且

attr_1是str_2等于1时metric_1列的值，

attr_2是str_2等于1时metric_1列的值，

.....

attr_1001 是 str_2 等于 1 时列 metric_2 的值

...

反规范化查询是：

 INSERT INTO denorm_table        
(dt, user, attr_1, attr_2, ..., attr_1000, attr_1001, attr_2000)
 SELECT 
'2018-11-01' as dt,
str_1,
arrayElement( groupArray(metric_1), indexOf(groupArray(str_2), '1') ) as attr_1,
arrayElement( groupArray(metric_1), indexOf(groupArray(str_2), '2') ) as attr_2,
......
arrayElement( groupArray(metric_1), indexOf(groupArray(str_2), '1000') ) as attr_1000,
arrayElement( groupArray(metric_2), indexOf(groupArray(str_2), '1001') ) as attr_1001,
.....
arrayElement( groupArray(metric_1), indexOf(groupArray(str_2), '2000') ) as attr_2000
FROM src_table
WHERE str_2 in ('1', '2', .....)
GROUP BY str_1

对于列 str_2 的 750 个值（非规范化表中的 1502 列），查询正常工作。

但是当非规范化表的列数为 2002（以及相应的 str_2 值 1000）时，我有一个 socket.timeout: timed out 错误

  File "/usr/lib/python2.7/site-packages/clickhouse_driver/client.py", line 119, in execute
columnar=columnar
  File "/usr/lib/python2.7/site-packages/clickhouse_driver/client.py", line 192, in process_ordinary_query
columnar=columnar)
  File "/usr/lib/python2.7/site-packages/clickhouse_driver/client.py", line 42, in receive_result
return result.get_result()
  File "/usr/lib/python2.7/site-packages/clickhouse_driver/result.py", line 39, in get_result
for packet in self.packet_generator:
  File "/usr/lib/python2.7/site-packages/clickhouse_driver/client.py", line 54, in packet_generator
packet = self.receive_packet()
  File "/usr/lib/python2.7/site-packages/clickhouse_driver/client.py", line 68, in receive_packet
packet = self.connection.receive_packet()
 File "/usr/lib/python2.7/site-packages/clickhouse_driver/connection.py", line 331, in receive_packet
packet.type = packet_type = read_varint(self.fin)
 File "/usr/lib/python2.7/site-packages/clickhouse_driver/reader.py", line 38, in read_varint
i = _read_one(f)
 File "/usr/lib/python2.7/site-packages/clickhouse_driver/reader.py", line 23, in _read_one
c = f.read(1)
 File "/usr/lib64/python2.7/socket.py", line 380, in read
data = self._sock.recv(left)
socket.timeout: timed out

客户端/服务器中是否有可以消除问题的设置？

---

来自日志

2018.12.04 22:49:26.755926 [ 36 ] {} <Trace> SystemLog (system.query_thread_log): Flushing system log
2018.12.04 22:49:26.756233 [ 139 ] {821ce7ea-94b7-4675-96f5-feccb31b0ebe} <Error> executeQuery: Code: 32, e.displayText() = DB::Exception: Attempt to read after eof, e.what() = DB::Exception (from [::1]:52224) (in query:

========= 编辑 =========

我修改了如下查询（@johey）并且错误没有再次发生：

WHERE modulo(sipHash64(str_1), 20) = 0 用于分组拆分数据 而不是对列 str_1 中的所有值运行查询

INSERT INTO dst_table (....)
SELECT
    arrayElement(metric_1_array, indexOf(str_2_array, '1') ) as attr_1,
    arrayElement(metric_1_array, indexOf(str_2_array, '2') ) as attr_2,
    ......
    arrayElement(metric_2_array, indexOf(str_2_array, '1') ) as attr_1001,
    ......
FROM (
    SELECT
        str_1,
        groupArray(metric_1) metric_1_array,
        groupArray(metric_2) metric_2_array,
        groupArray(str_2) str_2_array
    FROM src_table
    WHERE modulo(sipHash64(str_1), 20) = 0
    AND str_2 in ('1', '2', ......)
    GROUP BY str_1
)

Answer 1

看起来你已经炸毁了 AST 解析器。有两个设置可以限制输入 AST 树的大小：max_ast_elements 和 max_expanded_ast_elements。您可以尝试抬起它们，看看是否有帮助。

这是实现这种非规范化过程的另一种方法，

SELECT
    arrayStringConcat(arrayConcat(['2018-11-01', str_1], arraySlice(groupArray(toString(metric_1)), 1, 1000), arraySlice(groupArray(toString(metric_2)), 1001, 1000)), ',')
FROM
(
    SELECT *
    FROM src_table
    WHERE toUInt64(str_2) BETWEEN 1 AND 2000
    ORDER BY toUInt64(str_2)
)
GROUP BY str_1
INTO OUTFILE 'xxx.csv'

然后将该 csv 数据文件加载到denorm_table。

更新

这个PR被合并后，你可以像这样实现服务器端ETL：

INSERT INTO denorm_table
SELECT
    arrayStringConcat(arrayConcat(['2018-11-01', str_1], arraySlice(groupArray(toString(metric_1)), 1, 1000), arraySlice(groupArray(toString(metric_2)), 1001, 1000)), ',')
FROM
(
    SELECT *
    FROM src_table
    WHERE toUInt64(str_2) BETWEEN 1 AND 2000
    ORDER BY toUInt64(str_2)
)
GROUP BY str_1
FORMAT CSV