红色的写性能差(更新不是真正的多线程)

Question

(我已经回答了)我正在与大约100个用户的红宝石公司的糟糕表现作斗争。为了测试它，我设置了一个简单的场景来更新五个不同示例问题上的“已完成百分比”。所有更新都在同一秒钟内开始，使用curl。令我惊讶的是，redmine中的更新似乎是序列化的(我指的不是多线程)，其时间安排如下：

real    0m1.122s
real    0m1.404s
real    0m2.258s
real    0m2.782s
real    0m3.151s
real    0m3.336s

或者另一次尝试，当其他会话的某些处理也受到阻碍时：

real    0m7.492s
real    0m7.803s
real    0m8.045s
real    0m8.246s
real    0m8.597s
real    0m8.825s

在production.log中处理大量SQL查询后的关键时刻是提交阶段。注意提交是如何在毫秒内完成的，但是在提交之间发生了一些神秘的事情，不仅序列化它们，而且需要几秒钟才能完成状态。这段日志是完整的-我没有删除中间的任何行。

   (0.6ms)  SELECT COUNT(*) FROM `custom_workflows` WHERE `custom_workflows`.`active` = 1 AND `custom_workflows`.`observable` = 'issue' AND (is_for_all OR EXISTS (SELECT * FROM custom_workflows_projects WHERE project_id=123 AND custom_workflow_id=id))
   (30.1ms)  COMMIT
Redirected to http://redmine.mydomain.local/issues/11973
Completed 302 Found in 7418.4ms (ActiveRecord: 135.8ms)
   (0.6ms)  SELECT COUNT(*) FROM `custom_workflows` WHERE `custom_workflows`.`active` = 1 AND `custom_workflows`.`observable` = 'issue' AND (is_for_all OR EXISTS (SELECT * FROM custom_workflows_projects WHERE project_id=123 AND custom_workflow_id=id))
   (24.8ms)  COMMIT
Redirected to http://redmine.mydomain.local/issues/9240
Completed 302 Found in 7738.3ms (ActiveRecord: 57.4ms)
   (0.4ms)  SELECT COUNT(*) FROM `custom_workflows` WHERE `custom_workflows`.`active` = 1 AND `custom_workflows`.`observable` = 'issue' AND (is_for_all OR EXISTS (SELECT * FROM custom_workflows_projects WHERE project_id=123 AND custom_workflow_id=id))
   (25.9ms)  COMMIT
Redirected to http://redmine.mydomain.local/issues/9614
Completed 302 Found in 7949.0ms (ActiveRecord: 135.7ms)
   (0.5ms)  SELECT COUNT(*) FROM `custom_workflows` WHERE `custom_workflows`.`active` = 1 AND `custom_workflows`.`observable` = 'issue' AND (is_for_all OR EXISTS (SELECT * FROM custom_workflows_projects WHERE project_id=123 AND custom_workflow_id=id))
   (24.6ms)  COMMIT
Redirected to http://redmine.mydomain.local/issues/12016
Completed 302 Found in 8058.5ms (ActiveRecord: 102.5ms)
   (0.5ms)  SELECT COUNT(*) FROM `custom_workflows` WHERE `custom_workflows`.`active` = 1 AND `custom_workflows`.`observable` = 'issue' AND (is_for_all OR EXISTS (SELECT * FROM custom_workflows_projects WHERE project_id=123 AND custom_workflow_id=id))
   (21.2ms)  COMMIT
Redirected to http://redmine.mydomain.local/issues/8853
Completed 302 Found in 8472.5ms (ActiveRecord: 90.4ms)
   (0.5ms)  SELECT COUNT(*) FROM `custom_workflows` WHERE `custom_workflows`.`active` = 1 AND `custom_workflows`.`observable` = 'issue' AND (is_for_all OR EXISTS (SELECT * FROM custom_workflows_projects WHERE project_id=123 AND custom_workflow_id=id))
   (27.9ms)  COMMIT
Redirected to http://redmine.mydomain.local/issues/10007
Completed 302 Found in 8700.4ms (ActiveRecord: 137.4ms)

我已经查过了：

• 乘客状态有超过足够的线程。
• apache2 httpd有超过足够多的进程。
• mysql 5.1使用innodb
• 无名氏不显示锁
• innotop不过量I/O
• iostat没有显示过多的I/O (大约10-15 IOPS，磁盘最大使用时间为17% )
• top不会显示CPU使用率过高(最大的70%可能是400%)。

Answer 1

好的，它是SMTP序列化的；默认情况下，它是不被记录的。

我改变了/var/www/html/redmine/config/configuration.yml：

  email_delivery:
    # delivery_method: :smtp       # slows down everything!
    delivery_method: :async_smtp