Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-21925

Redo log mutex competition is serious

Details

    Description

      In the high concurrent write scenario, the redo log mutex takes up a lot of CPU time。
      In order to solve this problem,MySQL #WL10310 optimizes the redo log(unlock and write concurrently)。
      Does MariaDB have any improvement plans?

      Attachments

        Issue Links

          Activity

            MDEV-21534, which was implemented recently in 10.5, should reduce the mutex contention.

            You filed MDEV-21923 for a very similar question, which I answered there by linking to existing tickets.

            marko Marko Mäkelä added a comment - MDEV-21534 , which was implemented recently in 10.5, should reduce the mutex contention. You filed MDEV-21923 for a very similar question, which I answered there by linking to existing tickets.

            yaojiapeng, this is exactly what MDEV-21534 was for. You won't see as much log_write_up_to() CPU use after this.
            Also, high concurrency scenario is much more improved.

            wlad Vladislav Vaintroub added a comment - yaojiapeng , this is exactly what MDEV-21534 was for. You won't see as much log_write_up_to() CPU use after this. Also, high concurrency scenario is much more improved.

            The ib_logfile0 format change in MDEV-14425 enabled us to replace log_sys.mutex with a read-write latch log_sys.latch in MDEV-27774. Before that format change, log_sys.mutex was a serious contention point, because it not only covered the copying of local log snippets from mtr_t::m_log to the shared log_sys.buf, but it was also held while encrypting log blocks or computing checksums.

            marko Marko Mäkelä added a comment - The ib_logfile0 format change in MDEV-14425 enabled us to replace log_sys.mutex with a read-write latch log_sys.latch in MDEV-27774 . Before that format change, log_sys.mutex was a serious contention point, because it not only covered the copying of local log snippets from mtr_t::m_log to the shared log_sys.buf , but it was also held while encrypting log blocks or computing checksums.

            People

              Unassigned Unassigned
              yaojiapeng peng
              Votes:
              0 Vote for this issue
              Watchers:
              3 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.