QUERY_RESPONSE_TIME plugin

The slow query log provides exact information about queries that take a long time to execute. However, sometimes there are a large number of queries that each take a very short amount of time to execute. This feature provides a tool for analyzing that information by counting and displaying the number of queries according to the the length of time they took to execute. The user can define time intervals that divide the range 0 to positive infinity into smaller intervals and then collect the number of commands whose execution times fall into each of those intervals.

This feature is based on Percona Response Time Distribution.

Each interval is described as:

(range_base ^ n; range_base ^ (n+1)]

The range_base is some positive number (see Limitations). The interval is defined as the difference between two nearby powers of the range base.

For example, if the range base=10, we have the following intervals:

(0; 10 ^ -6], (10 ^ -6; 10 ^ -5], (10 ^ -5; 10 ^ -4], ..., (10 ^ -1; 10 ^1], (10^1; 10^2]...(10^7; positive infinity]

or

(0; 0.000001], (0.000001; 0.000010], (0.000010; 0.000100], ..., (0.100000; 1.0]; (1.0; 10.0]...(1000000; positive infinity]

For each interval, a count is made of the queries with execution times that fell into that interval.

You can select the range of the intervals by changing the range base. For example, for base range=2 we have the following intervals:

(0; 2 ^ -19], (2 ^ -19; 2 ^ -18], (2 ^ -18; 2 ^ -17], ..., (2 ^ -1; 2 ^1], (2 ^ 1; 2 ^ 2]...(2 ^ 25; positive infinity]

or

(0; 0.000001], (0.000001, 0.000003], ..., (0.25; 0.5], (0.5; 2], (2; 4]...(8388608; positive infinity]

Small numbers look strange (i.e., don’t look like powers of 2), because we lose precision on division when the ranges are calculated at runtime. In the resulting table, you look at the high boundary of the range.

For example, you may see:

MariaDB [test]> SELECT * FROM INFORMATION_SCHEMA.QUERY_RESPONSE_TIME;
+----------------+-------+----------------+
| TIME           | COUNT | TOTAL          |
+----------------+-------+----------------+
|       0.000001 |     0 |       0.000000 |
|       0.000010 |    17 |       0.000094 |
|       0.000100 |  4301         0.236555 |
|       0.001000 |  1499 |       0.824450 |
|       0.010000 | 14851 |      81.680502 |
|       0.100000 |  8066 |     443.635693 |
|       1.000000 |     0 |       0.000000 |
|      10.000000 |     0 |       0.000000 |
|     100.000000 |     1 |      55.937094 |
|    1000.000000 |     0 |       0.000000 |
|   10000.000000 |     0 |       0.000000 |
|  100000.000000 |     0 |       0.000000 |
| 1000000.000000 |     0 |       0.000000 |
| TOO LONG       |     0 | TOO LONG       |
+----------------+-------+----------------+

This means there were:

* 17 queries with 0.000001 < query execution time < = 0.000010 seconds; total execution time of the 17 queries = 0.000094 seconds

* 4301 queries with 0.000010 < query execution time < = 0.000100 seconds; total execution time of the 4301 queries = 0.236555 seconds

* 1499 queries with 0.000100 < query execution time < = 0.001000 seconds; total execution time of the 1499 queries = 0.824450 seconds

* 14851 queries with 0.001000 < query execution time < = 0.010000 seconds; total execution time of the 14851 queries = 81.680502 seconds

* 8066 queries with 0.010000 < query execution time < = 0.100000 seconds; total execution time of the 8066 queries = 443.635693 seconds

* 1 query with 10.000000 < query execution time < = 100.0000 seconds; total execution time of the 1 query = 55.937094 seconds

Usage

SELECT

You can get the distribution using the query:

SELECT * FROM INFORMATION_SCHEMA.QUERY_RESPONSE_TIME;

You can write a complex query like:

SELECT c.count, c.time,
(SELECT SUM(a.count) FROM INFORMATION_SCHEMA.QUERY_RESPONSE_TIME as a WHERE a.count != 0) as query_count,
(SELECT COUNT(*)     FROM INFORMATION_SCHEMA.QUERY_RESPONSE_TIME as b WHERE b.count != 0) as not_zero_region_count,
(SELECT COUNT(*)     FROM INFORMATION_SCHEMA.QUERY_RESPONSE_TIME) as region_count
FROM INFORMATION_SCHEMA.QUERY_RESPONSE_TIME as c WHERE c.count > 0;

Note: If query_response_time_stats is ON, the execution times for these two SELECT queries will also be collected.

SHOW

Unlike in original patch, SHOW QUERY_RESPONSE_TIME is not supported.

FLUSH

Unlike in original patch, FLUSH QUERY_RESPONSE_TIME is not supported. Flushing can be done with:

SET GLOBAL query_response_time_flush=1;

It does two things:

• Clears the collected times from INFORMATION_SCHEMA.QUERY_RESPONSE_TIME table
• Reads the value of query_response_time_range_base and uses it to set the range base for the table

Configuration

The plugin can be configured with the command-line options (in the my.cnf file or on the command line). The following command-line options are available: