Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-29508

perfschema.short_option_1 fails with MSAN - Error in accept: Invalid argument



    • Bug
    • Status: In Progress (View Workflow)
    • Minor
    • Resolution: Unresolved
    • 10.5
    • 10.5
    • Tests
    • None


      LD_LIBRARY_PATH=<path to MSAN libs> perl mysql-test-run.pl perfschema.short_option_1

      10.5 1ac0bce36e

      ***Warnings generated in error logs during shutdown after running tests: perfschema.short_option_1

      Version: '10.5.17-MariaDB-debug-log'  socket: '/home/jenkins/10.5/mysql-test/var/tmp/mysqld.1.sock'  port: 16000  Source distribution
      2022-09-11  1:02:33 0 [ERROR] Error in accept: Invalid argument
      2022-09-11  1:02:33 0 [Note] /home/jenkins/10.5/sql/mariadbd (initiated by: unknown): Normal shutdown
      2022-09-11  1:02:33 0 [Note] Event Scheduler: Purging the queue. 0 events
      2022-09-11  1:02:33 0 [Note] Debug sync points hit:                   1941
      2022-09-11  1:02:33 0 [Note] Debug sync points executed:              0
      2022-09-11  1:02:33 0 [Note] Debug sync points max active per thread: 0
      2022-09-11  1:02:33 0 [Note] /home/jenkins/10.5/sql/mariadbd: Shutdown complete
      User time 0.26, System time 0.18
      Maximum resident set size 351552, Integral resident set size 0
      Non-physical pagefaults 112789, Physical pagefaults 0, Swaps 0
      Blocks in 0 out 912, Messages in 0 out 0, Signals 0
      Voluntary context switches 64, Involuntary context switches 2

      My build was built with


      but it's also happening in the Foundation buildbot.
      I only observed it on 10.5, not sure about the buildbot (since it's a post-test failure, it's not stored in the cross-reference).

      The effect is deterministic at least on 10.5. It started happening after this patch

      commit 122742897b47a19c85b1a5e9932ab3a8c2a4134e
      Author: Sergei Golubchik
      Date:   Thu Jun 23 12:59:31 2022 +0200
          my_safe_process: try to kill the process softly first
          first SIGTERM and if the process didn't die in 10 seconds, SIGKILL it.
          This allows various tools like `rr`, `gcov`, `gprof`, etc to flush
          their data to disk properly

      I can't tell whether the change introduced a bogus failure or revealed a previously hidden problem.


        Issue Links



              monty Michael Widenius
              elenst Elena Stepanova
              0 Vote for this issue
              3 Start watching this issue



                Git Integration

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