Comment 1 for bug 1395249

Testcase setup (in my case it took about a night of runtime to trigger);

1. Attached tar is a recent version of sysbench. Extract in home dir, creates ~/sysbenchnew
2. Start server (cd basedir; ~/percona-qa/startup.sh 0; ./start)
3. Prep sysbech:
    export LD_PRELOAD=${PWD}/bundle/libmysqlclient_r.so.16
    ~/sysbenchnew/sysbench --num-threads=64 --report-interval=10 --max-time=0 --max-requests=0 --rand-type=pareto --oltp-table-size=1000000 --mysql-user=root --test=/home/roel/sysbenchnew/oltp.lua --mysql-socket=/sda/Percona-Server-5.6.21-rel70.0-676-debug.Linux.x86_64/socket.sock --mysql-db=test prepare
4. Put the following line which starts two sysbench runs at the same time (notice the &) in a script named "3"
   ~/sysbenchnew/sysbench --num-threads=64 --report-interval=10 --max-time=0 --max-requests=0 --rand-type=pareto --oltp-table-size=1000000 --mysql-user=root --test=/home/roel/sysbenchnew/oltp.lua --mysql-socket=/sda/Percona-Server-5.6.21-rel70.0-676-debug.Linux.x86_64/socket.sock --mysql-db=test run &
   ~/sysbenchnew/sysbench --num-threads=64 --report-interval=10 --max-time=0 --max-requests=0 --rand-type=pareto --oltp-table-size=1000000 --mysql-user=root --test=/home/roel/sysbenchnew/oltp.lua --mysql-socket=/sda/Percona-Server-5.6.21-rel70.0-676-debug.Linux.x86_64/socket.sock --threads=120 --mysql-db=test run &
5. Make a script named 4 and start it;
   $ cat sysbenchnew_scripts/4
   while(true); do
     ./3 &
     DELAY=$[1 + ( $RANDOM % ( 3600 ) ) ]
     sleep $DELAY
     kill -9 `pidof sysbench`
   done
6. Setup a light monitoring script (maybe not needed, just adding for completeness) and start it (all script in different shells);
   $ cat sysbenchnew_scripts/2
      while(true); do ./bin/mysql -uroot -S/sda/Percona-Server-5.6.21-rel70.0-676-debug.Linux.x86_64/socket.sock test -e"show global status like 'Innodb_row_lock_current_waits'; show global status like 'questions'; show global status like 'queries'; "; sleep 0.2; done
7. Wait a night orso for the crash to happen.