Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-17481

mariadb service won't shutdown when it's running and the OS datetime updated backwards

Details

    Description

      When I update the OS datetime *backwards* while `mariadb` is running, it causes `mariadb` to hang when I command it's service to stop.
      I've also noticed that this doesn't occur when I set the new OS datetime value to be grater than the current OS datetime, only backwards causes the hang.

      —

      Steps to reproduce: 

       
      		 
          sudo date --set="2018-10-04 20:00:00"
      		 
       
      		 
          sudo service mysql start
      		 
           * Starting MariaDB database server mysqld [ OK ]
      		 
       
      		 
          sudo service mysql stop
      		 
           * Stopping MariaDB database server mysqld [ OK ]
      		 
       
      		 
          sudo service mysql start
      		 
           * Starting MariaDB database server mysqld [ OK ]
      		 
       
      		 
          sudo date --set="2018-10-04 10:00:00"
      		 
       
      		 
          sudo service mysql stop
      		 
           * Stopping MariaDB database server mysqld

      > setting the date, starting the service, resetting the date -10 hours, stopping the service => hang

      Attachments

        Issue Links

          duplicates

          Bug - A problem which impairs or prevents the functions of the product. MDEV-20078 When setting back the system time while mysqld is running, shutdown will hang

          • Major - Major loss of function.
          • Closed
          is duplicated by

          Bug - A problem which impairs or prevents the functions of the product. MDEV-20598 Server slow to shutdown if system time goes backwards

          • Blocker - Blocks development and/or testing work, production could not run.
          • Closed

          Bug - A problem which impairs or prevents the functions of the product. MDEV-22789 MariaDB fails to stop on OS shutdown

          • Major - Major loss of function.
          • Closed
          relates to

          Task - A task that needs to be done. MDEV-16526 Overhaul the InnoDB page flushing

          • Major - Major loss of function.
          • Closed

          Activity

            Ascending order - Click to sort in descending order
            View 8 older comments
            marko Marko Mäkelä added a comment -

            wlad, what do you think of this hack? Basically, we would intentionally trigger some ‘spurious wakeups’ to work around the problem. Ultimately we want to rewrite the page cleaner (MDEV-16526) and get rid of the problematic call, but that would be a way too big change in GA releases.

            marko Marko Mäkelä added a comment - wlad , what do you think of this hack? Basically, we would intentionally trigger some ‘spurious wakeups’ to work around the problem. Ultimately we want to rewrite the page cleaner ( MDEV-16526 ) and get rid of the problematic call, but that would be a way too big change in GA releases.
            wlad Vladislav Vaintroub added a comment - - edited

            waking at shutdown seems fine, but do we really need it at other times, frequently? buf_flush_event is set here and there anyway, in different situations.

            wlad Vladislav Vaintroub added a comment - - edited waking at shutdown seems fine, but do we really need it at other times, frequently? buf_flush_event is set here and there anyway, in different situations.
            wlad Vladislav Vaintroub added a comment -

            https://jira.mariadb.org/browse/MDEV-17481?focusedCommentId=133652&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-133652
            probably causes a regression, because it uses nanoseconds timer my_interval_timer(), where previously milliseconds were used.

            ulint next_loop_time = ut_time_ms() + 1000; vs ulint next_loop_time = my_interval_timer() + 1000;

            is a big difference.

            wlad Vladislav Vaintroub added a comment - https://jira.mariadb.org/browse/MDEV-17481?focusedCommentId=133652&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-133652 probably causes a regression, because it uses nanoseconds timer my_interval_timer(), where previously milliseconds were used. ulint next_loop_time = ut_time_ms() + 1000; vs ulint next_loop_time = my_interval_timer() + 1000; is a big difference.
            marko Marko Mäkelä added a comment -

            wlad, the old comment that you refer to is not part of the currently planned solution. We would not increase the use of my_interval_timer(), but instead basically just ensure that a call to os_event_set(buf_flush_event) will end the infinite wait here and there.

            marko Marko Mäkelä added a comment - wlad , the old comment that you refer to is not part of the currently planned solution. We would not increase the use of my_interval_timer() , but instead basically just ensure that a call to os_event_set(buf_flush_event) will end the infinite wait here and there.
            marko Marko Mäkelä added a comment -

            The work-around looks OK to me. We’d be waking up the page cleaner in logs_empty_and_mark_files_at_shutdown() and srv_master_do_idle_tasks().

            marko Marko Mäkelä added a comment - The work-around looks OK to me. We’d be waking up the page cleaner in logs_empty_and_mark_files_at_shutdown() and srv_master_do_idle_tasks() .

            People

              thiru Thirunarayanan Balathandayuthapani
              jossef Jossef Harush
              Votes:
              0 Vote for this issue
              Watchers:
              11 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved:

                Git Integration

                  Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.