InnoDB: Failing assertion: purge_sys->state == PURGE_STATE_RUN | abort (sig=6) in innodb_sched_priority_purge_update

================== Use version:
Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug

================== Testcase:
SET GLOBAL INNODB_PURGE_STOP_NOW=ON;
SET GLOBAL innodb_sched_priority_purge=0;

================== Server start command:
[roel@localhost Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug]$ cat start
if [ -r /usr/lib64/libjemalloc.so.1 ]; then export LD_PRELOAD=/usr/lib64/libjemalloc.so.1
elif [ -r /usr/lib/x86_64-linux-gnu/libjemalloc.so.1 ]; then export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so.1
elif [ -r /sda/Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug/lib/mysql/libjemalloc.so.1 ]; then export LD_PRELOAD=/sda/Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug/lib/mysql/libjemalloc.so.1
else echo 'Error: jemalloc not found, please install it first'; exit 1; fi
/sda/Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug/bin/mysqld --innodb_buffer_pool_size=2147483648 --basedir=/sda/Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug --datadir=/sda/Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug/data --plugin-load=tokudb=ha_tokudb.so --core --socket=/sda/Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug/socket.sock --port=13781 --log-error=/sda/Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug/log/master.err 2>&1 &
echo 'Server socket: /sda/Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug/socket.sock with datadir: /sda/Percona-Server-5.6.20-rel68.0-657.Linux.x86_64-debug/data'

Maybe a sister bug of http://bugs.mysql.com/bug.php?id=73913 ?

Idem for SET GLOBAL innodb_sched_priority_purge=-1

A debug build-only facility for MTR testcases does not support this particular scenario: setting the purge thread priority while the purge is paused.

Another testcase

CREATE TABLE t2(s1 INT);
CREATE TABLE t1(c INT);
flush tables t1,t2 with read lock;
SET GLOBAL innodb_sched_priority_purge=0;

Roel -

Can you provide more info re. #5? First, the top three statements seem to be rather arbitrary in relation to the bug, second, running it does not fail.

Another testcase

DROP DATABASE test;CREATE DATABASE test;USE test;
CREATE TABLE t2(ticket int,inhalt text,KEY tig (ticket),fulltext index tix (inhalt)) ENGINE=InnoDB;
create table t1(t1 timestamp default now(),t2 DATE,t3 timestamp DEFAULT '0000-00-00 00:00:00');
FLUSH LOCAL TABLES test.t1,test.t2 for ExPoRt;
SET GLOBAL innodb_sched_priority_purge=19;

As mentioned elsewhere, please use CLI

mysql> \s
--------------
/sda/Percona-Server-5.6.21-rel69.0-681.Linux.x86_64-debug/bin/mysql Ver 14.14 Distrib 5.6.21-69.0, for Linux (x86_64) using EditLine wrapper

Connection id: 1
Current database: test
Current user: root@localhost
SSL: Not in use
Current pager: stdout
Using outfile: ''
Using delimiter: ;
Server version: 5.6.21-69.0-debug Percona Server (GPL), Release 69.0, Revision 681, DEBUG BINARY
Protocol version: 10
Connection: Localhost via UNIX socket
Server characterset: latin1
Db characterset: latin1
Client characterset: utf8
Conn. characterset: utf8
UNIX socket: /sda/Percona-Server-5.6.21-rel69.0-681.Linux.x86_64-debug/socket.sock
Uptime: 4 sec

Threads: 1 Questions: 5 Slow queries: 0 Opens: 67 Flush tables: 1 Open tables: 60 Queries per second avg: 1.250
--------------

mysql> CREATE TABLE t2(s1 INT);
sched_priority_purge=0;Query OK, 0 rows affected (0.21 sec)

mysql> CREATE TABLE t1(c INT);
Query OK, 0 rows affected (0.09 sec)

mysql> flush tables t1,t2 with read lock;
Query OK, 0 rows affected (0.06 sec)

mysql> SET GLOBAL innodb_sched_priority_purge=0;
ERROR 2013 (HY000): Lost connection to MySQL server during query

(Sorry, that was before realizing it was CLI vs MTR issue)

The difference between CLI and MTR here is extremely suspicious in the context of this bug.

The difference between CLI and MTR is that the latter runs with default-storage-engine=MyISAM by default. Change it on the command line or specify InnoDB explicitly in DDL, it will fail just fine.
Tried 5.6.21-69.0-debug-log

Thanks Elena

Thanks Elena, never looked into what option it was that triggered non-replay on all cli vs mtr issues (regular topic)

Ref bug 1519204

A much simpler test case to reproduce this problem
*******************************************
SET GLOBAL innodb_purge_stop_now = ON;
SET GLOBAL innodb_sched_priority_purge=0;
*******************************************

It looks like "innodb_sched_priority_purge_update()" does not take into account that "SET GLOBAL innodb_sched_priority_purge = xxx" can be called when purge threads are disabled.
Therefore, we run into the following assertion
"ut_ad(purge_sys->state == PURGE_STATE_RUN);"

https://github.com/percona/percona-server/pull/259
https://github.com/percona/percona-server/pull/260

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-1535