减少 MySQL RAM 使用量

Question

我目前在我的 16gb 服务器上为我的大约 15 个网站（主要是 WordPress 和 PHP 框架）使用以下配置：

key_buffer_size                = 256M
max_allowed_packet             = 256M
thread_stack                   = 256K
thread_cache_size              = 256K
table_open_cache               = 10000
table_definition_cache         = 8192

innodb_file_per_table          = 1
innodb_flush_log_at_trx_commit = 2
innodb_flush_method            = O_DIRECT
innodb_buffer_pool_size        = 10G
innodb_buffer_pool_instances   = 10
innodb_log_file_size           = 1G
innodb_read_io_threads         = 8
innodb_write_io_threads        = 8
innodb_io_capacity             = 2000

tmp_table_size                 = 256M
max_heap_table_size            = 256M

sort_buffer_size               = 4M
read_buffer_size               = 256K
join_buffer_size               = 256K
read_rnd_buffer_size           = 256K

query_cache_size               = 0
query_cache_type               = 0
query_cache_limit              = 128M

myisam-recover-options         = BACKUP
max_connections                = 500

几个月来一切都很顺利，但我发现优化和内存使用之间的完美平衡（有些网站真的是 SQL 密集型的）。但是几天前我看到 MySQL 被杀死了，因为 ram 使用量超过了可用 RAM（如果我记得的话，在 16 上超过 10 GB）。

所以我的问题是，当这种情况发生时，您应该首先在 MySQL 配置中减少什么？我在所有 *_buffer_size 和 inno_pool_size 之间犹豫不决，因此为了在更短的时间内释放更多 RAM，我将这些变量减少为：

innodb_buffer_pool_size        = 8G
innodb_buffer_pool_instances   = 8

目前一切都很好，我现在是 9gb/16gb，但我不知道这是否可行。

谢谢。

编辑：添加来自网站的示例查询

"SELECT c.cover, c.type, cl.slug, cl.label, GROUP_CONCAT(p.id) AS links
 FROM card c
 INNER JOIN card_language cl ON cl.card_id = c.id AND cl.language_id = " . $language->id . "
 INNER JOIN card_has_card_category chcc ON chcc.card_id = c.id
 INNER JOIN card_category_language ccl ON ccl.card_category_id = chcc.card_category_id AND ccl.language_id = " . $language->id . "
 LEFT JOIN card_has_links cp ON cp.card_id = c.id 
 LEFT JOIN link p ON p.id = cp.link_id
 WHERE c.type = '" . $type . "' 
 AND ccl.slug = '" . $slug . "'
 GROUP BY cl.slug
 ORDER BY c.id DESC
 LIMIT " . $limit . "
 OFFSET " . $offset .  "
 ;"

Answer 1

一般情况下，最好使用 MySQL 提供的默认值。

我确实看到您可能将某些设置增加到危险的数量。请记住，如果 MySQL 使用了太多需要交换的 RAM，性能可能会一落千丈。

key_buffer_size                = 256M  --> 20M (Assuming no tables are MyISAM)
max_allowed_packet             = 256M  -- no more than 1% of RAM
thread_stack                   = 256K  -- remove (to get default)
thread_cache_size              = 256K  -- This is NOT bytes!!; change to 20
table_open_cache               = 10000 -- Do you really have thousands of tables?
table_definition_cache         = 8192  -- Do you really have thousands of tables?

innodb_file_per_table          = 1
innodb_flush_log_at_trx_commit = 2
innodb_flush_method            = O_DIRECT
innodb_buffer_pool_size        = 10G  -- the main tunable; 10G is good
innodb_buffer_pool_instances   = 10
innodb_log_file_size           = 1G
innodb_read_io_threads         = 8
innodb_write_io_threads        = 8
innodb_io_capacity             = 2000  -- Good, IF you have SSD, not HDD

tmp_table_size                 = 256M  -- no more than 1% of RAM
max_heap_table_size            = 256M  -- no more than 1% of RAM

sort_buffer_size               = 4M
read_buffer_size               = 256K
join_buffer_size               = 256K
read_rnd_buffer_size           = 256K

query_cache_size               = 0
query_cache_type               = 0
query_cache_limit              = 128M

myisam-recover-options         = BACKUP
max_connections                = 500  -- see below

SHOW GLOBAL STATUS LIKE 'Max_used_connections'; 向max_connections 提供“高水位标记”。您可能不需要比“使用”的数字大得多的 `max_connections。

为了加速 WP，因此整个系统，O.Jones 和我制作了一个 WordPress 插件来改进某些 INDEXes: https://wordpress.org/plugins/index-wp-mysql-for-speed 我们检测各种东西，比如 InnoDB 存储的 Barrucuda 版本的存在引擎和其他 MySQL 奥秘，并做正确的事。