push_down.md

TiSpark push down

Overview

TiSpark can push down some operators to TiKV

• predicate
• aggregate
• sort
• limit
• topN

Here is the detail table:

data type	sum	count	avg	min	max	predicate & order by & group by
BIT	❌	✅	❌	✅	✅	❌
BOOLEAN	✅	✅	✅	✅	✅	✅
TINYINT	✅	✅	✅	✅	✅	✅
SMALLINT	✅	✅	✅	✅	✅	✅
MEDIUMINT	✅	✅	✅	✅	✅	✅
INTEGER	✅	✅	✅	✅	✅	✅
BIGINT	✅	✅	✅	✅	✅	✅
FLOAT	❌	✅	✅	✅	✅	✅
DOUBLE	✅	✅	✅	✅	✅	✅
DECIMAL	✅	✅	✅	✅	✅	✅
DATE	❌	✅	❌	✅	✅	✅
DATETIME	❌	✅	❌	✅	✅	✅
TIMESTAMP	❌	✅	❌	✅	✅	✅
TIME	❌	✅	❌	？	？	✅
YEAR	✅	✅	✅	✅	✅	✅
CHAR	❌	✅	❌	✅	✅	✅
VARCHAR	❌	✅	❌	✅	✅	✅
TINYTEXT	❌	✅	❌	✅	✅	✅
TEXT	❌	✅	❌	✅	✅	✅
MEDIUMTEXT	❌	✅	❌	✅	✅	✅
LONGTEXT	❌	✅	❌	✅	✅	✅
BINARY	❌	✅	❌	✅	✅	✅
VARBINARY	❌	✅	❌	✅	✅	✅
TINYBLOB	❌	✅	❌	✅	✅	❌
BLOB	❌	✅	❌	✅	✅	❌
MEDIUMBLOB	❌	✅	❌	✅	✅	❌
LONGBLOB	❌	✅	❌	✅	✅	❌
ENUM	❌	✅	❌	✅	✅	❌
SET	❌	❌	❌	？	？	❌

• min/max(time) will get wrong data now
• min/max(set) may cause TiKV panic now

Aggregate

Aggregate push down is supported since TiSpark 2.5.0

Here are some limits:

• Sum(float) can't be pushed down because Spark will cast float to double, but cast can't be pushed down in TiSpark.
• Count(set) can't be pushed down because TiKV does not support set type.
• Count(col1,col2,...) can't be pushed down because TiDB/TiKV does not support Count(col1,col2,...).
• Avg can be pushed down when both the sum and count can be pushed down.