MySQL Select and Sort Status Variables

This page is a reboot of the original written 15 years ago in 2005. Back then, I must have been using MySQL 4.1 or 5.0. Today, I am using MySQL 8.0.22. A lot changed in MySQL from 4.1 to 8.0, but it still has the same Select_% and Sort_% status variables which are equally important today in 2021. Let’s examine them in greater detail and with better examples.

Check out my book to learn a lot more about MySQL performance:

Efficient MySQL Performance book cover

The MySQL Select_% and Sort_% status variables (metrics) are:

  • Select_scan
  • Select_range
  • Select_full_join
  • Select_full_range_join
  • Select_range_check
  • Sort_scan
  • Sort_range
  • Sort_merge_passes
  • Sort_rows

Using MySQL 8.0.22 Community, the test tables are:

CREATE TABLE `t1` (
  `id` int NOT NULL AUTO_INCREMENT,
  `c1` int NOT NULL,
  `c2` int NOT NULL,
  PRIMARY KEY (`id`),
  KEY `c1` (`c1`)
) ENGINE=InnoDB

CREATE TABLE `t2` (
  `id` int NOT NULL AUTO_INCREMENT,
  `c1` int NOT NULL,
  `c2` int NOT NULL,
  PRIMARY KEY (`id`),
  KEY `c1` (`c1`)
) ENGINE=InnoDB

Each table has about 30 rows.

Select

Metric First Table Join Table
Select_scan
Select_range
Select_full_join
Select_full_range_join
Select_range_check

Select_scan and Select_range apply to the first (or only) table. Select_full_join, Select_full_range_join, and Select_range_check apply to the second and subsequent tables.

MySQL table join order is determined by the query planner (to create the best execution plan) not the JOIN clause clause (unless STRAIGHT_JOIN is used).

mysql> EXPLAIN SELECT * FROM t1, t2 WHERE t1.c1 = t2.c2 ORDER BY t1.c2;
+-------------+-------+------+---------------+------+---------+------------+------+----------+---------------------------------+
| select_type | table | type | possible_keys | key  | key_len | ref        | rows | filtered | Extra                           |
+-------------+-------+------+---------------+------+---------+------------+------+----------+---------------------------------+
| SIMPLE      | t2    | ALL  | NULL          | NULL | NULL    | NULL       |   26 |   100.00 | Using temporary; Using filesort |
| SIMPLE      | t1    | ref  | c1            | c1   | 4       | test.t2.c2 |    1 |   100.00 | NULL                            |
+-------------+-------+------+---------------+------+---------+------------+------+----------+---------------------------------+

mysql> EXPLAIN FORMAT=TREE SELECT * FROM t1, t2 WHERE t1.c1 = t2.c2 ORDER BY t1.c2;
+-------------------------------------------------------------------------+
| EXPLAIN                                                                 |
+-------------------------------------------------------------------------+
| -> Sort: t1.c2
    -> Stream results  (cost=11.95 rows=26)
        -> Nested loop inner join  (cost=11.95 rows=26)
            -> Table scan on t2  (cost=2.85 rows=26)
            -> Index lookup on t1 using c1 (c1=t2.c2)  (cost=0.25 rows=1) |
+-------------------------------------------------------------------------+

Top output is a standard EXPLAIN plan with the first table on top, so t2 is the first table and t1 is the second table. Note that the tables are listed differently in the FROM clause. MySQL is accessing table t2 first because it produces a better execution plan which you can see in detail by tracing the optimizer.

Bottom output is an “EXPLAIN tree” which is new as of MySQL 8.0. It shows the query execution plan from the bottom up, per-node. The EXPLAIN tree reads:

  1. Nested loop inner join
    1. Table scan on t2
    2. Index lookup on t1 using c1 (c1=t2.c2)
  2. Stream results
  3. Sort: t1.c2

Step 3 refers to “Using temporary” in the EXPLAIN plan, and step 4 refers to “Using filesort”. See How MySQL executes ORDER BY by Sergei Petrunia for a beautifully illustrated explanation of MySQL ORDER BY.

The first table, t2, is a table scan (“type: ALL”), so Select_scan will be incremented. The second table, t1, is neither a table scan nor a range scan, so no metric will be incremented because there are no select metrics for other join types. Let’s clear the session metrics, execute the query, and check the select metrics:

mysql> FLUSH STATUS;

mysql> SELECT * FROM t1, t2 WHERE t1.c1 = t2.c2 ORDER BY t1.c2;
--
-- Output removed
--
20 rows in set (0.00 sec)

mysql> SHOW SESSION STATUS LIKE 'Select_%';
+------------------------+-------+
| Variable_name          | Value |
+------------------------+-------+
| Select_full_join       | 0     |
| Select_full_range_join | 0     |
| Select_range           | 0     |
| Select_range_check     | 0     |
| Select_scan            | 1     | -- t2
+------------------------+-------+

As expected, Select_scan = 1 due to the table scan on t2.

Select_scan

Select_scan refers to the first table in a join, or the only table if there is only one table. It means MySQL did a table or index scan. For such tables, EXPLAIN lists “ALL” or “index” in the “type” column.

Table scans (“type: ALL”) are usually (but not always) terrible for performance because reading an entire table is slow. However, table scans are common. It’s not uncommon to see a server where 50% of all SELECT queries are Select_scan. A SELECT results in a table scan when no index can be used to for the WHERE conditions. From a performance perspective it’s safe to say that you always want to decrease Select_scan—zero is best. However, in some cases the value of Select_scan can increase after optimizing queries because MySQL is able to do more. Ultimately, Select_scan is a performance limiter which should decrease to allow increased overall QPS (queries per second).

Index scans (“type: index”) are better than table scans, especially when “Extra: Using index” appears in the EXPLAIN plan which means a covering index is being used. Without “Using index”, the query is still a full table scan but MySQL uses the index to read rows in order, which is usually the result of an ORDER BY optimization.

Select_range

Select_range refers to the first table in a join, or the only table if there is only one table. It means MySQL read a range of rows from the table. For such tables, EXPLAIN lists “type: range”. MySQL used an index to limit table reads to rows within the range, which saves time that would otherwise be wasted on disk seeks for non-matching rows. Therefore, Select_range is a lot faster than Select_scan. A query like SELECT * FROM t1 WHERE c1 > 5 AND c1 < 13 specifies an exclusive range between 5 and 13.. Since column c1 is indexed, MySQL reads the index for the range to find and read only rows in that range. If c1 was not indexed, MySQL would have to do a full table scan. See Range Optimization for a lot more details.

Select_full_join

Select_full_join refers to the second or greater table in a join. It means MySQL did a table scan to join the table to the preceding table. For such tables, EXPLAIN lists “ALL” in the “type” column. Select_full_join results when no index can be used to join the table. Like Select_scan, this is a performance limiter that should be avoided.

The worst case is a table scan on both join tables:

mysql> EXPLAIN SELECT * FROM t1, t2 WHERE t1.c2 = t2.c2 ;
+-------------+-------+------+---------------+------+---------+------+------+----------+--------------------------------------------+
| select_type | table | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra                                      |
+-------------+-------+------+---------------+------+---------+------+------+----------+--------------------------------------------+
| SIMPLE      | t1    | ALL  | NULL          | NULL | NULL    | NULL |   20 |   100.00 | NULL                                       |
| SIMPLE      | t2    | ALL  | NULL          | NULL | NULL    | NULL |   26 |    10.00 | Using where; Using join buffer (hash join) |
+-------------+-------+------+---------------+------+---------+------+------+----------+--------------------------------------------+

mysql> EXPLAIN FORMAT=TREE SELECT * FROM t1, t2 WHERE t1.c2 = t2.c2 ;
+-------------------------------------------------------------------------+
| EXPLAIN                                                                 |
+-------------------------------------------------------------------------+
| -> Inner hash join (t2.c2 = t1.c2)  (cost=54.50 rows=52)
    -> Table scan on t2  (cost=0.03 rows=26)
    -> Hash
        -> Table scan on t1  (cost=2.25 rows=20)                          |
+-------------------------------------------------------------------------+

mysql> SHOW SESSION STATUS LIKE 'Select_%';
+------------------------+-------+
| Variable_name          | Value |
+------------------------+-------+
| Select_full_join       | 1     | -- t2
| Select_full_range_join | 0     |
| Select_range           | 0     |
| Select_range_check     | 0     |
| Select_scan            | 1     | -- t1
+------------------------+-------+

The “type: ALL” for both tables indicates a table scan. Select_full_join was incremented for the table scan on the second table (t2), and Select_scan was incremented for the table scan on the first table (t1). This query does table scans because the tables are joined on column c2 which is not indexed on either table.

This is the worst case because, as the MySQL manual notes:

Because type is ALL for each table, this output indicates that MySQL is generating a Cartesian product of all the tables; that is, every combination of rows. This takes quite a long time, because the product of the number of rows in each table must be examined.

The first table has 20 rows and the second has 26 rows, so the result is 20 * 26 = 520 rows. The worst I have ever seen was a three table join that caused a cross product of 112 billion rows (actually, the query never finished before it caused the server to halt). With only 1,000 rows in each table, the product is 1 million rows which will be noticeably slow.

Both EXPLAIN outputs show another new feature in MySQL 8.0: hash joins. See Hash join in MySQL 8 by the MySQL Server Team for an illustrated explanation.

Select_full_range_join

Select_full_range_join refers to the second or greater table in a join. It means MySQL joined the table to the preceding table using a range scan, which is significantly better than Select_full_join. For such tables, EXPLAIN lists “type: range”. Like Select_range, Select_full_range_join requires an index and the same range optimizations apply.

mysql> EXPLAIN SELECT * FROM t1, t2 WHERE t1.c1 = 10 AND t2.c1 > 13;
+-------------+-------+-------+---------------+------+---------+-------+------+----------+------------------------------------------------------+
| select_type | table | type  | possible_keys | key  | key_len | ref   | rows | filtered | Extra                                                |
+-------------+-------+-------+---------------+------+---------+-------+------+----------+------------------------------------------------------+
| SIMPLE      | t1    | ref   | c1            | c1   | 4       | const |    1 |   100.00 | NULL                                                 |
| SIMPLE      | t2    | range | c1            | c1   | 4       | NULL  |   13 |   100.00 | Using index condition; Using join buffer (hash join) |
+-------------+-------+-------+---------------+------+---------+-------+------+----------+------------------------------------------------------+

mysql> EXPLAIN FORMAT=TREE SELECT * FROM t1, t2 WHERE t1.c1 = 10 AND t2.c1 > 13;
+-----------------------------------------------------------------------------------------------+
| EXPLAIN                                                                                                                                                                                                                                     |
+-----------------------------------------------------------------------------------------------+
| -> Inner hash join (no condition)  (cost=6.46 rows=13)
    -> Index range scan on t2 using c1, with index condition: (t2.c1 > 13)  (cost=6.11 rows=13)
    -> Hash
        -> Index lookup on t1 using c1 (c1=10)  (cost=0.35 rows=1)                              |
+-----------------------------------------------------------------------------------------------+

mysql> SHOW STATUS LIKE 'Select_%';
+------------------------+-------+
| Variable_name          | Value |
+------------------------+-------+
| Select_full_join       | 0     |
| Select_full_range_join | 1     | -- t2
| Select_range           | 0     |
| Select_range_check     | 0     |
| Select_scan            | 0     |
+------------------------+-------+

Note that the condition on t1 is constant: t1.c1 = 10. Since MySQL knows column values from the preceding table (t1), it also knows that it can use those column values to range scan the join table (t2). This is not always the case, as the next select metric indicates.

Select_range_check

Select_range_check refers to the second or greater table in a join. It means “MySQL checks whether it is possible to use a range or index_merge access method to retrieve rows.” In other words, it means MySQL wants to do Select_full_range_join but cannot because it does not know the column values from the preceding table, so it checks the index on the join table for each new row (column values) from the preceding table.

mysql> EXPLAIN SELECT * FROM t1, t2 WHERE t1.c1 > t2.c1;
+-------------+-------+------+---------------+------+---------+------+------+----------+------------------------------------------------+
| select_type | table | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra                                          |
+-------------+-------+------+---------------+------+---------+------+------+----------+------------------------------------------------+
| SIMPLE      | t1    | ALL  | c1            | NULL | NULL    | NULL |   20 |   100.00 | NULL                                           |
| SIMPLE      | t2    | ALL  | c1            | NULL | NULL    | NULL |   26 |    33.33 | Range checked for each record (index map: 0x2) |
+-------------+-------+------+---------------+------+---------+------+------+----------+------------------------------------------------+

mysql> EXPLAIN FORMAT=TREE SELECT * FROM t1, t2 WHERE t1.c1 > t2.c1;
+--------------------------------------------------------------------------------------+
| EXPLAIN                                                                              |
+--------------------------------------------------------------------------------------+
| -> Nested loop inner join  (cost=54.50 rows=173)
    -> Table scan on t1  (cost=2.25 rows=20)
    -> Filter: (t1.c1 > t2.c1)  (cost=0.06 rows=9)
        -> Index range scan on t2 (re-planned for each iteration)  (cost=0.06 rows=26) |
+--------------------------------------------------------------------------------------+

mysql> SHOW SESSION STATUS LIKE 'Select_%';
+------------------------+-------+
| Variable_name          | Value |
+------------------------+-------+
| Select_full_join       | 0     |
| Select_full_range_join | 0     |
| Select_range           | 0     |
| Select_range_check     | 1     | -- t2
| Select_scan            | 1     | -- t1
+------------------------+-------+

The condition t1.c1 > t2.c1 triggers the range check. As MySQL scans t1, it reads column c1 values. In this example, we know the values ahead of time because the test table is tiny and static. But in a real database, table writes and transaction isolation levels mean the values can change, therefore MySQL cannot know them ahead of time. For each row, MySQL checks if can use the index on t2.c1 to read the t2 row matching the current t1.c1 value. It is possible that t1.c1 has values not in t2.c1.

Even if MySQL determines that it can do a range scan (or index merge), it does not increment Select_full_range_join. Only Select_range_check is incremented.

See “Range checked for each record” in EXPLAIN Output Format for a good explanation.

Sort

The sort metrics Sort_scan, Sort_range, Sort_merge_passes, and Sort_rows apply to queries that use ORDER BY or GROUP BY, regardless of how many tables.

A sort as indicated by “filesort” in EXPLAIN output. The term “filesort” is misleading. The MySQL Server Team says, “In MySQL, filesort is the catch-all algorithm for producing sorted results for ORDER-BY or GROUP-BY queries.” (Filesort optimization in 5.7.3: pack values in the sort buffer) See also What does Using filesort mean in MySQL?

Memory allocation for sorting is set by the system variable sort_buffer_size. Resist the temptation to “tune” this variable or simply increase its value. Read How to Tune MySQL’s Sort_buffer_size by Baron and More on understanding sort_buffer_size by Ronald—both are long-time MySQL experts you can trust. Also read Impact of the sort buffer size in MySQL by Percona.

A of MySQL 8.0.12, sort_buffer_size is allocated incrementally as needed.
Before MySQL 8.0.12, the full size is allocated up front.

For a deeper technical understanding of sorting in MySQL 8.0, read Sorting in The Unofficial MySQL 8.0 Optimizer Guide by Morgan—another long-time MySQL expert you can trust.

Sort_scan

As of MySQL 8.0, Sort_scan is the total number of all sort operations. In MySQL 5.7 and older, Sort_scan was the number of sort operations for queries not using a range scan; see Sort_range.

The value of Sort_scan is usually very high because sorting rows is common. Monitor the rate of this metric (sort ops/second) and watch for unexpected changes.

Sort_range

As of MySQL 8.0, Sort_range means nothing. The status variable remains, but the code that incremented it was removed. In MySQL 5.7 the code was in sql/filesort.cc:

  if (tab->quick())
    thd->inc_status_sort_range();
  else
    thd->inc_status_sort_scan();

But in MySQL 8.0 that if-else block was replaced with a single line:

  thd->inc_status_sort_scan();

In MySQL 5.7 and older, Sort_range and Sort_scan were mutually exclusive, as shown in the code snippet above. Sort_range was incremented instead of Sort_scan for queries using a range scan.

`Select_scan`  + filesort = `Sort_scan`++
`Select_range` + filesort = `Sort_range`++

Let’s use MySQL 5.7.31 and the salaries table in the employees sample database:

CREATE TABLE `salaries` (
  `emp_no` int(11) NOT NULL,
  `salary` int(11) NOT NULL,
  `from_date` date NOT NULL,
  `to_date` date NOT NULL,
  PRIMARY KEY (`emp_no`,`from_date`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4

Column salary is not indexed, so the following query causes a table scan, not a range scan, which means Sort_scan will be incremented:

mysql> EXPLAIN SELECT emp_no, salary FROM salaries WHERE salary > 250000 ORDER BY emp_no, salary LIMIT 10;
+-------------+----------+------------+------+---------------+------+---------+------+---------+----------+-----------------------------+
| select_type | table    | partitions | type | possible_keys | key  | key_len | ref  | rows    | filtered | Extra                       |
+-------------+----------+------------+------+---------------+------+---------+------+---------+----------+-----------------------------+
| SIMPLE      | salaries | NULL       | ALL  | NULL          | NULL | NULL    | NULL | 2834690 |    33.33 | Using where; Using filesort |
+-------------+----------+------------+------+---------------+------+---------+------+---------+----------+-----------------------------+

mysql> SELECT emp_no, salary FROM salaries WHERE salary > 250000 ORDER BY emp_no, salary LIMIT 10;
Empty set (0.74 sec)

mysql> SHOW SESSION STATUS LIKE 'Sort_%';
+-------------------+-------+
| Variable_name     | Value |
+-------------------+-------+
| Sort_merge_passes | 0     |
| Sort_range        | 0     |
| Sort_rows         | 0     |
| Sort_scan         | 1     | -- Due to type: ALL
+-------------------+-------+

Now the proof: index salary to cause a range scan:

mysql> ALTER TABLE salaries ADD INDEX (salary);
Query OK, 0 rows affected (15.52 sec)
Records: 0  Duplicates: 0  Warnings: 0

mysql> EXPLAIN SELECT emp_no, salary FROM salaries WHERE salary > 250000 ORDER BY emp_no, salary LIMIT 10;
+-------------+----------+------------+-------+---------------+--------+---------+------+------+----------+------------------------------------------+
| select_type | table    | partitions | type  | possible_keys | key    | key_len | ref  | rows | filtered | Extra                                    |
+-------------+----------+------------+-------+---------------+--------+---------+------+------+----------+------------------------------------------+
| SIMPLE      | salaries | NULL       | range | salary        | salary | 4       | NULL |    1 |   100.00 | Using where; Using index; Using filesort |
+-------------+----------+------------+-------+---------------+--------+---------+------+------+----------+------------------------------------------+

mysql> FLUSH STATUS;

mysql> SELECT emp_no, salary FROM salaries WHERE salary > 250000 ORDER BY emp_no, salary LIMIT 10;
Empty set (0.18 sec)

mysql> SHOW SESSION STATUS LIKE 'Sort_%';
+-------------------+-------+
| Variable_name     | Value |
+-------------------+-------+
| Sort_merge_passes | 0     |
| Sort_range        | 1     | -- Due to type: range
| Sort_rows         | 0     |
| Sort_scan         | 0     |
+-------------------+-------+

Therefore, in MySQL 5.7 and older, the total number of all sort operations is Sort_scan + Sort_range.

Sort_merge_passes

Sort_merge_passes is the number of merge buffer operations. Sort_merge_passes is incremented by 1 each time function merge_buffers() is called (in sql/filesort.cc).

When sort_buffer_size is not large enough to hold all unsorted rows, MySQL writes “merge buffers” to temporary files on disk, then merges the buffers to produce the final, sorted result.

Let’s use a table with 1 million rows created by sysbench:

CREATE TABLE `sbtest1` (
  `id` int NOT NULL AUTO_INCREMENT,
  `k` int NOT NULL DEFAULT '0',
  `c` char(120) NOT NULL DEFAULT '',
  `pad` char(60) NOT NULL DEFAULT '',
  PRIMARY KEY (`id`),
  KEY `k_1` (`k`)
) ENGINE=InnoDB AUTO_INCREMENT=1000001 DEFAULT CHARSET=utf8 |

The query,

SELECT c FROM sbtest1 WHERE id > 100 ORDER BY c

produces a sorted result of 999,900 rows which does not fit in any sane sort_buffer_size, thereby causing merge buffers to disk and sort merge passes.

As mentioned at the beginning of the Sort section: Resist the temptation to “tune” this variable or simply increase its value. But to demonstrate how sort_buffer_size affects Sort_merge_passes and query response time, let’s try the query above with different sort buffer sizes: 32 KB (minimum), 10 MB (very large), 32 MB (dangerous), and 128 MB (insane).

sort_buffer_size = 32 KB (minimum)

mysql> SELECT c FROM sbtest1 WHERE id > 100 ORDER BY c;
--
-- Output removed
--
999900 rows in set (3.60 sec)

mysql> SHOW SESSION STATUS LIKE 'Sort_%';
+-------------------+--------+
| Variable_name     | Value  |
+-------------------+--------+
| Sort_merge_passes | 1917   |
| Sort_range        | 0      |
| Sort_rows         | 999900 |
| Sort_scan         | 1      |
+-------------------+--------+
"filesort_summary": {
  "memory_available": 32768,
  "key_size": 240,
  "row_size": 610,
  "max_rows_per_buffer": 53,
  "num_rows_estimate": 493200,
  "num_rows_found": 999900,
  "num_initial_chunks_spilled_to_disk": 11494,
  "peak_memory_used": 33368,
  "sort_algorithm": "std::sort",
  "sort_mode": "<fixed_sort_key, packed_additional_fields>"
}

With only 32 KB, MySQL must use 11,494 merge buffers as indicated by num_initial_chunks_spilled_to_disk in the optimizer trace output at bottom. (This value was fixed as of MySQL 8.0.2: “In optimizer trace output, num_tmp_files did not actually indicate number of files. It has been renamed to num_initial_chunks_spilled_to_disk and indicates the number of chunks before any merging has occurred. (Bug #25733784, Bug #85487)”.) The number of merge passes (1,917) is far less because each call to merge_buffers() merges several buffers (currently 7: constexpr size_t MERGEBUFF = 7; in sql/sql_sort.h). All 32 KB was used, as indicated by peak_memory_used.

The query took 3.6s to execute.

sort_buffer_size = 10 MB (very large)

Does a very large sort buffer increase query response time? Let’s try 10 MB:

mysql> SELECT c FROM sbtest1 WHERE id > 100 ORDER BY c;
--
-- Output removed
--
999900 rows in set (2.49 sec)

mysql> SHOW SESSION STATUS LIKE 'Sort_%';
+-------------------+--------+
| Variable_name     | Value  |
+-------------------+--------+
| Sort_merge_passes | 6      |
| Sort_range        | 0      |
| Sort_rows         | 999900 |
| Sort_scan         | 1      |
+-------------------+--------+
"filesort_summary": {
  "memory_available": 10485760,
  "key_size": 240,
  "row_size": 610,
  "max_rows_per_buffer": 17189,
  "num_rows_estimate": 493200,
  "num_rows_found": 999900,
  "num_initial_chunks_spilled_to_disk": 36,
  "peak_memory_used": 10486104,
  "sort_algorithm": "std::stable_sort",
  "sort_mode": "<fixed_sort_key, packed_additional_fields>"
}

Yes, query response time decreased 30% to 2.49s due to a 99.7% decrease in Sort_merge_passes: 1917 to 6. MySQL used all 10 MB of the sort buffer.

sort_buffer_size = 32 MB (dangerous)

mysql> SELECT c FROM sbtest1 WHERE id > 100 ORDER BY c;
--
-- Output removed
--
999900 rows in set (2.24 sec)

mysql> SHOW SESSION STATUS LIKE 'Sort_%';
+-------------------+--------+
| Variable_name     | Value  |
+-------------------+--------+
| Sort_merge_passes | 1      |
| Sort_range        | 0      |
| Sort_rows         | 999900 |
| Sort_scan         | 1      |
+-------------------+--------+
"filesort_summary": {
  "memory_available": 33554432,
  "key_size": 240,
  "row_size": 610,
  "max_rows_per_buffer": 55007,
  "num_rows_estimate": 493200,
  "num_rows_found": 999900,
  "num_initial_chunks_spilled_to_disk": 12,
  "peak_memory_used": 33557336,
  "sort_algorithm": "std::stable_sort",
  "sort_mode": "<fixed_sort_key, packed_additional_fields>"
}

With a 32 MB sort buffer, merge passes drops to 1 but query response time remains similar to a 10 MB sort buffer: 2.24s vs. 2.49s, respectively. Not surprising: 1 merge pass vs. 6 is not a significant difference.

I call a 32 MB sort_buffer_size “dangerous” for two reasons. First, sort buffers are allocated per-connection, so it can use a lot of memory on busy servers. Second, several MySQL experts recommend against large values. Granted, each server is different, so do your own analysis and use a larger sort_buffer_size if there is a proven need for it. As of MySQL 8.0.12, sort_buffer_size is allocated incrementally as needed, so it is less dangerous.

sort_buffer_size = 128 MB (insane)

mysql> SELECT c FROM sbtest1 WHERE id > 100 ORDER BY c;
--
-- Output removed
--
999900 rows in set (2.52 sec)

mysql> SHOW SESSION STATUS LIKE 'Sort_%';
+-------------------+--------+
| Variable_name     | Value  |
+-------------------+--------+
| Sort_merge_passes | 1      |
| Sort_range        | 0      |
| Sort_rows         | 999900 |
| Sort_scan         | 1      |
+-------------------+--------+
"filesort_summary": {
  "memory_available": 134217728,
  "key_size": 240,
  "row_size": 610,
  "max_rows_per_buffer": 220029,
  "num_rows_estimate": 493200,
  "num_rows_found": 999900,
  "num_initial_chunks_spilled_to_disk": 3,
  "peak_memory_used": 134229360,
  "sort_algorithm": "std::stable_sort",
  "sort_mode": "<fixed_sort_key, packed_additional_fields>"
}

A 128 MB sort_buffer_size proves an important point: performance can decrease when sort_buffer_size is too large.

Query response time was 2.52s with 128 MB compared to 2.24s with 32 MB and 2.49s with 10 MB. Why? I do not have formal proof, but I am confident that it is due to the fact that Sort_merge_passes is negligible (1, 1, 6, respectively) but it takes longer to allocate 128 MB since, as of MySQL 8.0.12, sort buffer memory is allocated incrementally as needed which means more overhead to allocate large amounts of memory.

Sort_rows

Sort_rows is the total number of rows sorted. The value is usually so high that it is nearly meaningless, even as a rate (rows sorted/second). If necessary, monitor the rate and watch for unexpected changes.