问题
I have a MySQL database table that's quite large with 2.5million rows and growing. To speed up queries I've added an index to one of columns. When I set the index manually, through PHPMyAdmin for example, it's cardinality is around 1500 which seems about right and my queries run without issue.
The problem then comes after a few queries, (especially on an INSERT but not limited to), have been run, the cardinality of that index drops to 17 or 18, and queries run extremely slow. Sometimes it seems to work it's way back to about 1500 or I have to do it through PHPMyAdmin again.
Is there any way to stop this cardinality drop from happening?
CREATE TABLE IF NOT EXISTS `probe_results` (
`probe_result_id` int(11) NOT NULL AUTO_INCREMENT,
`date` date NOT NULL,
`month` int(11) NOT NULL,
`year` int(11) NOT NULL,
`time` time NOT NULL,
`type` varchar(11) NOT NULL,
`probe_id` varchar(50) NOT NULL,
`status` varchar(11) NOT NULL,
`temp_1` decimal(11,0) NOT NULL,
`temp_2` decimal(11,0) NOT NULL,
`crc` varchar(11) NOT NULL,
`raw_data` text NOT NULL,
`txt_file` text NOT NULL,
PRIMARY KEY (`probe_result_id`),
KEY `probe_id` (`probe_id`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1 AUTO_INCREMENT=2527300 ;
The 'probe_result_id' column is the primary key, the probe_id is the column with the index in question.
Example query:
SELECT IF(b.reactive_total IS NULL, 0, b.reactive_total) AS reactive_total, a.* FROM (SELECT COUNT(CASE WHEN asset_testing_results.asset_testing_year = '2016' AND asset_testing_results.asset_testing_month = '7' AND asset_testing_results.asset_stopped = '0' AND asset_testing_results.asset_testing_completed = '0' THEN 1 END) AS due_total, (COUNT(CASE WHEN asset_testing_results.asset_testing_year = '2016' AND asset_testing_results.asset_stopped = '0' AND asset_testing_results.asset_testing_completed = '1' AND asset_testing_results.asset_testing_satisfactory = '1' AND asset_testing_results.asset_testing_actioned = '0' THEN 1 END)+(IF(probes_passed_total IS NULL, 0, probes_passed_total))) AS passed_total, (COUNT(CASE WHEN asset_testing_results.asset_testing_year = '2016' AND asset_testing_results.asset_stopped = '0' AND asset_testing_results.asset_testing_completed = '1' AND asset_testing_results.asset_testing_satisfactory = '0' AND asset_testing_results.asset_testing_actioned = '0' THEN 1 END)+(IF(probes_failed_total IS NULL, 0, probes_failed_total))) AS failed_total, COUNT(CASE WHEN asset_testing_results.asset_testing_year = '2016' AND asset_testing_results.asset_stopped = '0' AND asset_testing_results.asset_testing_completed = '1' AND asset_testing_results.asset_testing_actioned = '1' THEN 1 END) AS actioned_total, COUNT(CASE WHEN asset_testing_results.asset_testing_year = '2016' AND asset_testing_results.asset_testing_month < '7' AND asset_testing_results.asset_testing_completed = '0' AND asset_testing_results.asset_testing_satisfactory = '0' AND asset_testing_results.asset_stopped = '0' THEN 1 END) AS missed_total, site.site_key, site.site_name FROM site LEFT JOIN location ON location.site_key = site.site_key LEFT JOIN sub_location ON sub_location.location_key = location.location_key LEFT JOIN asset ON asset.sub_location_key = sub_location.sub_location_key AND asset.stopped = '0' LEFT JOIN asset_testing ON asset_testing.asset_type_key = asset.asset_type_key AND asset_testing.probe_assessed = '0' LEFT JOIN asset_testing_results ON asset_testing_results.asset_testing_key = asset_testing.asset_testing_key AND asset_testing_results.asset_key = asset.asset_key LEFT JOIN (SELECT site.site_key, COUNT(CASE WHEN p.probe_id IS NOT NULL AND p.asset_testing_key IS NOT NULL THEN 1 END) AS probes_passed_total, COUNT(CASE WHEN p.probe_id IS NOT NULL AND p.asset_testing_key IS NULL AND p.temp_1 IS NOT NULL THEN 1 END) AS probes_failed_total FROM assetsvs_probes LEFT JOIN (SELECT q.probe_id, q.month, q.year, IF(r.temp_1 IS NULL, q.temp_1, r.temp_1) as temp_1, r.asset_testing_key FROM (SELECT DISTINCT probe_results.probe_id, probe_results.month, probe_results.year, probe_results.temp_1 FROM probe_results LEFT JOIN assetsvs_probes ON assetsvs_probes.probe_id = probe_results.probe_id LEFT JOIN asset ON asset.asset_key = assetsvs_probes.asset_key LEFT JOIN sub_location ON sub_location.sub_location_key = asset.sub_location_key LEFT JOIN location ON location.location_key = sub_location.location_key LEFT JOIN site ON site.site_key = location.site_key WHERE site.client_key = '25')q LEFT JOIN (SELECT probe_results.month, probe_results.year, probe_results.probe_id, temp_1, asset_testing.asset_testing_key FROM probe_results LEFT JOIN assetsvs_probes ON assetsvs_probes.probe_id = probe_results.probe_id LEFT JOIN asset_testing ON asset_testing.asset_testing_key = assetsvs_probes.asset_testing_key LEFT JOIN asset ON asset.asset_key = assetsvs_probes.asset_key LEFT JOIN sub_location ON sub_location.sub_location_key = asset.sub_location_key LEFT JOIN location ON location.location_key = sub_location.location_key LEFT JOIN site ON site.site_key = location.site_key WHERE temp_1 != 'invalid' AND ((temp_1 >= test_min AND test_max = '') OR (temp_1 <= test_max AND test_min = '') OR (temp_1 >= test_min AND temp_1 <= test_max)) AND year = '2016' AND site.client_key = '25' GROUP BY probe_results.month, probe_results.year, probe_results.probe_id)r ON r.probe_id = q.probe_id AND r.month = q.month AND r.year = q.year WHERE q.year = '2016' GROUP BY probe_id, month, year) p ON p.probe_id = assetsvs_probes.probe_id LEFT JOIN asset_testing ON asset_testing.asset_testing_key = assetsvs_probes.asset_testing_key LEFT JOIN asset ON asset.asset_key = assetsvs_probes.asset_key LEFT JOIN sub_location ON sub_location.sub_location_key = asset.sub_location_key LEFT JOIN location ON location.location_key = sub_location.location_key LEFT JOIN site ON site.site_key = location.site_key GROUP BY site.site_key) probe_results ON probe_results.site_key = site.site_key WHERE site.client_key = '25' GROUP BY site.site_key)a LEFT JOIN (SELECT COUNT(CASE WHEN jobs.status = '3' THEN 1 END) AS reactive_total, site.site_key FROM jobs LEFT JOIN jobs_meta ON jobs_meta.job_id = jobs.job_id AND jobs_meta.meta_key = 'start_date' LEFT JOIN site ON site.site_key = jobs.site_key WHERE site.client_key = '25' AND jobs_meta.meta_value LIKE '%/2016 %' GROUP BY site.site_key)b ON b.site_key = a.site_key
Thanks
回答1:
Cardinality (along with other statistics) is calculated and updated by MySQL automatically, so you do not have direct means to prevent it from dropping.
However, you can take a few steps to make this less likely to happen or correct the behaviour.
First of all, MySQL updates index statistics for all supported table engines if you run analyze table command.
For innodb table engine MySQL provides a set of configuration settings that can influence the behaviour of the sampling. The settings and their effect are described in the MySQL documentation:
- non-persistent statistics
- persistent statistics
The main setting is innodb_stats_transient_sample_pages:
• Small values like 1 or 2 can result in inaccurate estimates of cardinality.
• Increasing the innodb_stats_transient_sample_pages value might require more disk reads. Values much larger than 8 (say, 100), can cause a significant slowdown in the time it takes to open a table or execute SHOW TABLE STATUS.
• The optimizer might choose very different query plans based on different estimates of index selectivity
.
For myisam MySQL dos not provide such a variety of settings. myisam_stats_method setting is described in the general index statistics documentation
来源：https://stackoverflow.com/questions/38302435/mysql-table-index-cardinality