Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-13908

10.1.27 Segfaulting

Details

    • Bug
    • Status: Closed (View Workflow)
    • Blocker
    • Resolution: Fixed
    • 10.1.27
    • 10.1.28
    • Galera
    • None
    • Ubuntu Xenial

    Description

      CI gate jobs that were passing this morning with 10.1.26 are now failing as my mirrors pulled in 10.1.27 apt packages.

      Log is attached or available at http://osa-ci.objects-us-dfw-1.cloud.lstn.net/454450/8/626/logs/openstack/aio1_galera_container-9fad60d2/mysql_logs/galera_server_error.log.txt (navigate up in the URI path to view the rest of the contents of /var/log in this container such as apt/* for package versions). ie. http://osa-ci.objects-us-dfw-1.cloud.lstn.net/454450/8/626/logs/openstack/aio1_galera_container-9fad60d2/apt/history.log.txt

      Attachments

        1. cluster.cnf
          1 kB
        2. debian.cnf
          0.3 kB
        3. mariadb.cnf
          0.4 kB
        4. my.cnf
          2 kB
        5. temp.txt
          12 kB
        6. tokudb.cnf
          0.3 kB

        Issue Links

          duplicates

          Bug - A problem which impairs or prevents the functions of the product. MDEV-13910 after update not starting

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

          Bug - A problem which impairs or prevents the functions of the product. MDEV-13787 Crash in persistent stats wsrep_on (thd=0x0)

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

          Activity

            Ascending order - Click to sort in descending order
            View 9 older comments
            anikitin Andrii Nikitin (Inactive) added a comment - - edited

            it really looks that drop table crashes only if table was created with innodb_stats_persistent=1 . InnoDB still tries to clear persistent stats for that table if it has any, despite innodb_stats_persistent=0 .

            So, to work around the problem:
            together with disabling persistent stats tables innodb*stats must be truncated:

            set global innodb_stats_persistent=0;
            		 
            truncate table mysql.innodb_index_stats;
            		 
            truncate table mysql.innodb_table_stats;

            anikitin Andrii Nikitin (Inactive) added a comment - - edited it really looks that drop table crashes only if table was created with innodb_stats_persistent=1 . InnoDB still tries to clear persistent stats for that table if it has any, despite innodb_stats_persistent=0 . So, to work around the problem: together with disabling persistent stats tables innodb*stats must be truncated: set global innodb_stats_persistent=0; truncate table mysql.innodb_index_stats; truncate table mysql.innodb_table_stats;
            rudyz Rudy Z added a comment -

            The amount of people this bug affected and the number of crashed servers is alarming, especially since Galera is used to operate clusters for fault-tolerance and load-balancing, and this bug could bring down a whole cluster, especially if small. Can there be more tests before a global release happens? I have seen major sites, including several I operate, going down altogether yesterday!

            rudyz Rudy Z added a comment - The amount of people this bug affected and the number of crashed servers is alarming, especially since Galera is used to operate clusters for fault-tolerance and load-balancing, and this bug could bring down a whole cluster, especially if small. Can there be more tests before a global release happens? I have seen major sites, including several I operate, going down altogether yesterday!
            fstrankowski Florian Strankowski added a comment -

            I agree with Rudy at this point. Its alarming that such a bug made it into the stable tree. Either QA failed at this point or is non-existent. It seems that 10.1.27 got revoked as of today and is not longer announced on official repos, way to late to be honest. Please do not ship such obvious faulty packages into stable tree - this is what the dev-tree is made for, appreciated.

            fstrankowski Florian Strankowski added a comment - I agree with Rudy at this point. Its alarming that such a bug made it into the stable tree. Either QA failed at this point or is non-existent. It seems that 10.1.27 got revoked as of today and is not longer announced on official repos, way to late to be honest. Please do not ship such obvious faulty packages into stable tree - this is what the dev-tree is made for, appreciated.
            aajii Antti-Jussi Korjonen added a comment -

            Anybody updated to 10.1.28 yet?

            aajii Antti-Jussi Korjonen added a comment - Anybody updated to 10.1.28 yet?
            anikitin Andrii Nikitin (Inactive) added a comment -

            10.1.28 got MDEV-13950 , where mysqld_safe (if you use it) may fail to start galera node properly and requires small patch to mysqld_safe .

            anikitin Andrii Nikitin (Inactive) added a comment - 10.1.28 got MDEV-13950 , where mysqld_safe (if you use it) may fail to start galera node properly and requires small patch to mysqld_safe .

            People

              anikitin Andrii Nikitin (Inactive)
              logan Logan V
              Votes:
              3 Vote for this issue
              Watchers:
              10 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.