过了这么久，discuz论坛的问题还是困扰着很多网友，其实从各论坛里看到的问题总结出来，很关键的一点都是因为没有将数据表引擎转成InnoDB导致的，discuz在并发稍微高一点的环境下就表现的非常糟糕，产生大量的锁等待，这时候如果把数据表引擎改成InnoDB的话，我相信会好很多。这次就写个扫盲贴吧。
1. 启用innodb引擎，并配置相关参数
#skip-innodb
innodb_additional_mem_pool_size = 16M #一般16M也够了，可以适当调整下 innodb_buffer_pool_size = 6G #如果是专用db的话，一般是内存总量的80% innodb_data_file_path = ibdata1:1024M:autoextend innodb_file_io_threads = 4 innodb_thread_concurrency = 20 innodb_flush_log_at_trx_commit = 1 innodb_log_buffer_size = 16M innodb_log_file_size = 256M innodb_log_files_in_group = 3 innodb_max_dirty_pages_pct = 50 innodb_lock_wait_timeout = 120 innodb_file_per_table
修改表引擎为innodb：
mysql> alter table cdb_access engine = innodb;
其他表类似上面，把表名换一下即可...
将表存储引擎改成innodb后，不仅可以避免大量的锁等待，还可以提升查询的效率，因为innodb会把data和index都放在buffer pool中，效率更高。
2.缓存优化
在 my.cnf 中添加/修改以下选项:
#取消文件系统的外部锁 skip-locking #不进行域名反解析,注意由此带来的权限/授权问题 skip-name-resolve #索引缓存,根据内存大小而定,如果是的db服务器,可以设置高达80%的内存总量 key_buffer = 512M #连接排队列表总数 back_log = 200 max_allowed_packet = 2M #打开表缓存总数,可以避免频繁的打开数据表产生的开销 table_cache = 512 #每个线程排序所需的缓冲 sort_buffer_size = 4M #每个线程读取索引所需的缓冲 read_buffer_size = 4M #MyISAM表发生变化时重新排序所需的缓冲 myisam_sort_buffer_size = M #缓存可重用的线程数 thread_cache = 128 #查询结果缓存 query_cache_size = 128M #设置超时时间,能避免长连接 set-variable = wait_timeout=60 #最大并发线程数,cpu数量*2 thread_concurrency = 4 #记录慢查询,然后对慢查询一一优化 log-slow-queries = slow.log long_query_time = 1 #关闭不需要的表类型,如果你需要,就不要加上这个 skip-bdb
以上参数根据各自服务器的配置差异进行调整,仅作为参考.
3.索引优化
上面提到了,已经开启了慢查询,那么接下来就要对慢查询进行逐个优化了.
搜索的查询SQL大致如下:
SELECT t.* FROM cdb_posts p, cdb_threads t WHERE
t.fid IN ('37', '45', '4', '6', '17', '41', '28', '32', '31', '1', '42')
AND p.tid=t.tid AND p.author LIKE 'JoansWin'
GROUP BY t.tid ORDER BY lastpost DESC LIMIT 0, 80;
用 EXPLAIN 分析的结果如下:
mysql>EXPLAIN SELECT t.* FROM cdb_posts p, cdb_threads t WHERE
t.fid IN ('37', '45', '4', '6', '17', '41', '28', '32', '31', '1', '42')
AND p.tid=t.tid AND p.author LIKE 'JoansWin'
GROUP BY t.tid ORDER BY lastpost DESC LIMIT 0, 80;
+-----------+------------+----------+--------------+-------------+-----------+-------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra +-----------+------------+----------+--------------+-------------+-----------+-------------+ | 1 | SIMPLE | t | range | PRIMARY,fid | fid | 2 | NULL | 66160 | Using where; Using temporary; Using filesort | | 1 | SIMPLE | p | ref | tid | tid | 3 | Forum.t.tid | 10 | Using where | +----+-------------+-------+-------+---------------+------+---------+-------------+-------+ ---------
只用到了 t.fid 和 p.tid,而 p.author 则没有索引可用,总共需要扫描
66160*10 = 661600 次索引,够夸张吧 :(
再分析 cdb_threads 和 cdb_posts 的索引情况:
mysql>show index from cdb_posts;
+-----------+------------+----------+--------------+-------------+-----------+---------- ---+----------+--------+------+--+ | Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | +-----------+------------+----------+--------------+---- ---------+-----------+-------------+----------+--------+------+--+ | cdb_posts | 0 | PRIMARY | 1 | pid | A | 680114 | NULL | NULL | | BTREE | | | cdb_posts | 1 | fid | 1 | fid | A | 10 | NULL | NULL | | BTREE | | | cdb_posts | 1 | tid | 1 | tid | A | 68011 | NULL | NULL | | BTREE | | | cdb_posts | 1 | tid | 2 | dateline | A | 680114 | NULL | NULL | | BTREE | | | cdb_posts | 1 | dateline | 1 | dateline | A | 680114 | NULL | NULL | | BTREE | | +-----------+------------+----------+--------------+-------------+-----------+---
以及
mysql>show index from cdb_threads;
+-----------+------------+----------+--------------+-------------+-----------+-------------+ ----------+--------+------+-----+ | Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | +-----------+------------+----------+--------------+----- --------+-----------+-------------+----------+--------+------+-----+ | cdb_threads | 0 | PRIMARY | 1 | tid | A | 68480 | NULL | NULL | | BTREE | | | cdb_threads | 1 | lastpost | 1 | topped | A | 4 | NULL | NULL | | BTREE | | | cdb_threads | 1 | lastpost | 2 | lastpost | A | 68480 | NULL | NULL | | BTREE | | | cdb_threads | 1 | lastpost | 3 | fid | A | 68480 | NULL | NULL | | BTREE | | | cdb_threads | 1 | replies | 1 | replies | A | 233 | NULL | NULL | | BTREE | | | cdb_threads | 1 | dateline | 1 | dateline | A | 68480 | NULL | NULL | | BTREE | | | cdb_threads | 1 | fid | 1 | fid | A | 10 | NULL | NULL | | BTREE | | | cdb_threads | 1 | enablehot | 1 | enablehot | A | 2 | NULL | NULL | | BTREE | | +-------------+------------+-----------+--------------+-------------+------
看到索引 fid 和 enablehot 基数太小,看来该索引完全没必要,不过,对于fid基数较大的情况,则可能需要保留>该索引.
所做修改如下:
ALTER TABLE `cdb_threads` DROP INDEX `enablehot`, DROP INDEX `fid`, ADD INDEX (`fid`, `lastpost`); ALTER TABLE `cdb_posts` DROP INDEX `fid`, ADD INDEX (`author`(10)); OPTIMIZE TABLE `cdb_posts`; OPTIMIZE TABLE `cdb_threads`;
在这里, p.author 字段我设定的部分索引长度是 10, 是我经过分析后得出来的结果,不同的系统,这里的长度也不同,最好自己先取一下平均值,然后再适当调整.
现在,再来执行一次上面的慢查询,发现时间已经从 6s 变成 0.19s,提高了 30 倍.