Explain Statements Using Aggregation
When aggregating data, the SQL Optimizer will select either a Hash Aggregation or Stream Aggregation operator. To improve query efficiency, aggregation is performed at both the coprocessor and TiDB layers. Consider the following example:
CREATE TABLE t1 (id INT NOT NULL PRIMARY KEY auto_increment, pad1 BLOB, pad2 BLOB, pad3 BLOB);
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM dual;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
INSERT INTO t1 SELECT NULL, RANDOM_BYTES(1024), RANDOM_BYTES(1024), RANDOM_BYTES(1024) FROM t1 a JOIN t1 b JOIN t1 c LIMIT 10000;
SELECT SLEEP(1);
ANALYZE TABLE t1;
From the output of SHOW TABLE REGIONS, you can see that this table is split into multiple Regions:
SHOW TABLE t1 REGIONS;
+-----------+--------------+--------------+-----------+-----------------+-------+------------+---------------+------------+----------------------+------------------+
| REGION_ID | START_KEY | END_KEY | LEADER_ID | LEADER_STORE_ID | PEERS | SCATTERING | WRITTEN_BYTES | READ_BYTES | APPROXIMATE_SIZE(MB) | APPROXIMATE_KEYS |
+-----------+--------------+--------------+-----------+-----------------+-------+------------+---------------+------------+----------------------+------------------+
| 64 | t_64_ | t_64_r_31766 | 65 | 1 | 65 | 0 | 1325 | 102033520 | 98 | 52797 |
| 66 | t_64_r_31766 | t_64_r_63531 | 67 | 1 | 67 | 0 | 1325 | 72522521 | 104 | 78495 |
| 68 | t_64_r_63531 | t_64_r_95296 | 69 | 1 | 69 | 0 | 1325 | 0 | 104 | 95433 |
| 2 | t_64_r_95296 | | 3 | 1 | 3 | 0 | 1501 | 0 | 81 | 63211 |
+-----------+--------------+--------------+-----------+-----------------+-------+------------+---------------+------------+----------------------+------------------+
4 rows in set (0.00 sec)
Using EXPLAIN with the following aggregation statement, you can see that └─StreamAgg_8 is first performed on each Region inside TiKV. Each TiKV Region will then send one row back to TiDB, which aggregates the data from each Region in StreamAgg_16:
EXPLAIN SELECT COUNT(*) FROM t1;
+----------------------------+-----------+-----------+---------------+---------------------------------+
| id | estRows | task | access object | operator info |
+----------------------------+-----------+-----------+---------------+---------------------------------+
| StreamAgg_16 | 1.00 | root | | funcs:count(Column#7)->Column#5 |
| └─TableReader_17 | 1.00 | root | | data:StreamAgg_8 |
| └─StreamAgg_8 | 1.00 | cop[tikv] | | funcs:count(1)->Column#7 |
| └─TableFullScan_15 | 242020.00 | cop[tikv] | table:t1 | keep order:false |
+----------------------------+-----------+-----------+---------------+---------------------------------+
4 rows in set (0.00 sec)
This is easiest to observe in EXPLAIN ANALYZE, where the actRows matches the number of Regions from SHOW TABLE REGIONS because a TableFullScan is being used and there are no secondary indexes:
EXPLAIN ANALYZE SELECT COUNT(*) FROM t1;
+----------------------------+-----------+---------+-----------+---------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+---------------------------------+-----------+------+
| id | estRows | actRows | task | access object | execution info | operator info | memory | disk |
+----------------------------+-----------+---------+-----------+---------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+---------------------------------+-----------+------+
| StreamAgg_16 | 1.00 | 1 | root | | time:12.609575ms, loops:2 | funcs:count(Column#7)->Column#5 | 372 Bytes | N/A |
| └─TableReader_17 | 1.00 | 4 | root | | time:12.605155ms, loops:2, cop_task: {num: 4, max: 12.538245ms, min: 9.256838ms, avg: 10.895114ms, p95: 12.538245ms, max_proc_keys: 31765, p95_proc_keys: 31765, tot_proc: 48ms, rpc_num: 4, rpc_time: 43.530707ms, copr_cache_hit_ratio: 0.00} | data:StreamAgg_8 | 293 Bytes | N/A |
| └─StreamAgg_8 | 1.00 | 4 | cop[tikv] | | proc max:12ms, min:12ms, p80:12ms, p95:12ms, iters:122, tasks:4 | funcs:count(1)->Column#7 | N/A | N/A |
| └─TableFullScan_15 | 242020.00 | 121010 | cop[tikv] | table:t1 | proc max:12ms, min:12ms, p80:12ms, p95:12ms, iters:122, tasks:4 | keep order:false | N/A | N/A |
+----------------------------+-----------+---------+-----------+---------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+---------------------------------+-----------+------+
4 rows in set (0.01 sec)
Hash Aggregation
The Hash Aggregation algorithm uses a hash table to store intermediate results while performing aggregation. It executes in parallel using multiple threads but consumes more memory than Stream Aggregation.
The following is an example of the HashAgg operator:
EXPLAIN SELECT /*+ HASH_AGG() */ count(*) FROM t1;
+---------------------------+-----------+-----------+---------------+---------------------------------+
| id | estRows | task | access object | operator info |
+---------------------------+-----------+-----------+---------------+---------------------------------+
| HashAgg_9 | 1.00 | root | | funcs:count(Column#6)->Column#5 |
| └─TableReader_10 | 1.00 | root | | data:HashAgg_5 |
| └─HashAgg_5 | 1.00 | cop[tikv] | | funcs:count(1)->Column#6 |
| └─TableFullScan_8 | 242020.00 | cop[tikv] | table:t1 | keep order:false |
+---------------------------+-----------+-----------+---------------+---------------------------------+
4 rows in set (0.00 sec)
The operator info shows that the hashing function used to aggregate the data is funcs:count(1)->Column#6.
Stream Aggregation
The Stream Aggregation algorithm usually consumes less memory than Hash Aggregation. However, this operator requires that data is sent ordered so that it can stream and apply the aggregation on values as they arrive.
Consider the following example:
CREATE TABLE t2 (id INT NOT NULL PRIMARY KEY, col1 INT NOT NULL);
INSERT INTO t2 VALUES (1, 9),(2, 3),(3,1),(4,8),(6,3);
EXPLAIN SELECT /*+ STREAM_AGG() */ col1, count(*) FROM t2 GROUP BY col1;
Query OK, 0 rows affected (0.11 sec)
Query OK, 5 rows affected (0.01 sec)
Records: 5 Duplicates: 0 Warnings: 0
+------------------------------+----------+-----------+---------------+---------------------------------------------------------------------------------------------+
| id | estRows | task | access object | operator info |
+------------------------------+----------+-----------+---------------+---------------------------------------------------------------------------------------------+
| Projection_4 | 8000.00 | root | | test.t2.col1, Column#3 |
| └─StreamAgg_8 | 8000.00 | root | | group by:test.t2.col1, funcs:count(1)->Column#3, funcs:firstrow(test.t2.col1)->test.t2.col1 |
| └─Sort_13 | 10000.00 | root | | test.t2.col1 |
| └─TableReader_12 | 10000.00 | root | | data:TableFullScan_11 |
| └─TableFullScan_11 | 10000.00 | cop[tikv] | table:t2 | keep order:false, stats:pseudo |
+------------------------------+----------+-----------+---------------+---------------------------------------------------------------------------------------------+
5 rows in set (0.00 sec)
In this example, the └─Sort_13 operator can be eliminated by adding an index on col1. Once the index is added, the data can be read in order and the └─Sort_13 operator is eliminated:
ALTER TABLE t2 ADD INDEX (col1);
EXPLAIN SELECT /*+ STREAM_AGG() */ col1, count(*) FROM t2 GROUP BY col1;
Query OK, 0 rows affected (0.28 sec)
+------------------------------+---------+-----------+----------------------------+----------------------------------------------------------------------------------------------------+
| id | estRows | task | access object | operator info |
+------------------------------+---------+-----------+----------------------------+----------------------------------------------------------------------------------------------------+
| Projection_4 | 4.00 | root | | test.t2.col1, Column#3 |
| └─StreamAgg_14 | 4.00 | root | | group by:test.t2.col1, funcs:count(Column#4)->Column#3, funcs:firstrow(test.t2.col1)->test.t2.col1 |
| └─IndexReader_15 | 4.00 | root | | index:StreamAgg_8 |
| └─StreamAgg_8 | 4.00 | cop[tikv] | | group by:test.t2.col1, funcs:count(1)->Column#4 |
| └─IndexFullScan_13 | 5.00 | cop[tikv] | table:t2, index:col1(col1) | keep order:true, stats:pseudo |
+------------------------------+---------+-----------+----------------------------+----------------------------------------------------------------------------------------------------+
5 rows in set (0.00 sec)
Multidimensional data aggregation with ROLLUP
Starting from v7.4.0, the GROUP BY clause of TiDB supports the WITH ROLLUP modifier.
In the GROUP BY clause, you can specify one or more columns as a group list and append the WITH ROLLUP modifier after the list. Then, TiDB will conduct multidimensional descending grouping based on the columns in the group list and provide you with summary results for each group in the output.
explain SELECT year, month, grouping(year), grouping(month), SUM(profit) AS profit FROM bank GROUP BY year, month WITH ROLLUP;
+----------------------------------------+---------+--------------+---------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| id | estRows | task | access object | operator info |
+----------------------------------------+---------+--------------+---------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| TableReader_44 | 2.40 | root | | MppVersion: 2, data:ExchangeSender_43 |
| └─ExchangeSender_43 | 2.40 | mpp[tiflash] | | ExchangeType: PassThrough |
| └─Projection_8 | 2.40 | mpp[tiflash] | | Column#6->Column#12, Column#7->Column#13, grouping(gid)->Column#14, grouping(gid)->Column#15, Column#9->Column#16 |
| └─Projection_38 | 2.40 | mpp[tiflash] | | Column#9, Column#6, Column#7, gid |
| └─HashAgg_36 | 2.40 | mpp[tiflash] | | group by:Column#6, Column#7, gid, funcs:sum(test.bank.profit)->Column#9, funcs:firstrow(Column#6)->Column#6, funcs:firstrow(Column#7)->Column#7, funcs:firstrow(gid)->gid, stream_count: 8 |
| └─ExchangeReceiver_22 | 3.00 | mpp[tiflash] | | stream_count: 8 |
| └─ExchangeSender_21 | 3.00 | mpp[tiflash] | | ExchangeType: HashPartition, Compression: FAST, Hash Cols: [name: Column#6, collate: binary], [name: Column#7, collate: utf8mb4_bin], [name: gid, collate: binary], stream_count: 8 |
| └─Expand_20 | 3.00 | mpp[tiflash] | | level-projection:[test.bank.profit, <nil>->Column#6, <nil>->Column#7, 0->gid],[test.bank.profit, Column#6, <nil>->Column#7, 1->gid],[test.bank.profit, Column#6, Column#7, 3->gid]; schema: [test.bank.profit,Column#6,Column#7,gid] |
| └─Projection_16 | 3.00 | mpp[tiflash] | | test.bank.profit, test.bank.year->Column#6, test.bank.month->Column#7 |
| └─TableFullScan_17 | 3.00 | mpp[tiflash] | table:bank | keep order:false, stats:pseudo |
+----------------------------------------+---------+--------------+---------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
10 rows in set (0.05 sec)
According to the GROUP BY year, month WITH ROLLUP syntax in the preceding statement, the SQL aggregation results for this statement can be calculated and concatenated in three groups: {year, month}, {year}, and {} respectively.
For more information, see GROUP BY modifiers.