Set mysql-show_processlist_extended make Proxysql crash

[Tusamarco]

• A clear description of the issue
  While running light load (5 threads 2 insert 2 select 1 delete ), once the parameter mysql-show_processlist_extended is inserted and loaded to runtime :
  update global_variables set Variable_Value=1 where Variable_name='mysql-show_processlist_extended';

ProxySQL crash.

/usr/bin/proxysql(_Z13crash_handleri+0x1a)[0x4b86da]
/lib64/libc.so.6(+0x363b0)[0x7ff9b30553b0]
/lib64/libc.so.6(+0x3af97)[0x7ff9b3059f97]
/usr/bin/proxysql(_ZN8Variable28fill_client_internal_sessionERN8nlohmann10basic_jsonISt3mapSt6vectorSsblmdSaNS0_14adl_serializerEEEi+0x5d7)[0x63b3b7]
/usr/bin/proxysql(_ZN13MySQL_Session39generate_proxysql_internal_session_jsonERN8nlohmann10basic_jsonISt3mapSt6vectorSsblmdSaNS0_14adl_serializerEEE+0x945)[0x513485]
/usr/bin/proxysql(_ZN21MySQL_Threads_Handler16SQL3_ProcesslistEv+0x6b6)[0x502486]
/usr/bin/proxysql(_ZN14ProxySQL_Admin25stats___mysql_processlistEv+0x4d)[0x539d3d]
/usr/bin/proxysql(_ZN14ProxySQL_Admin24GenericRefreshStatisticsEPKcjb+0x8d0)[0x55a1a0]
/usr/bin/proxysql(_Z21admin_session_handlerP13MySQL_SessionPvP10_PtrSize_t+0x14d)[0x5639cd]
/usr/bin/proxysql(_ZN13MySQL_Session7handlerEv+0x1a78)[0x5185c8]
/usr/bin/proxysql(_Z11child_mysqlPv+0x34d)[0x5320dd]
/lib64/libpthread.so.0(+0x7e65)[0x7ff9b4235e65]
/lib64/libc.so.6(clone+0x6d)[0x7ff9b311d88d]
2020-06-07 11:56:13 main.cpp:1564:ProxySQL_daemonize_phase3(): [ERROR] ProxySQL crashed. Restarting!

• ProxySQL version
  [root@proxysql1 proxysql]# proxysql --version
  ProxySQL version 2.0.13-31-g6f11ea6, codename Truls

• OS version
  Linux proxysql1 3.10.0-1062.12.1.el7.x86_64 Remove argument MySQL_Data_Stream from MySQL_Protocol's function #1 SMP Tue Feb 4 23:02:59 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux
  [root@proxysql1 proxysql]# cat /etc/redhat-release
  CentOS Linux release 7.7.1908 (Core)

• The steps to reproduce the issue
  For me was enough to change the parameter, see attached file for query digest

• The full ProxySQL error log (default location: /var/lib/proxysql/proxysql.log)
  see attached compressed file

If this is a crashing bug, please also include:

• The package used to install ProxySQL
  See attached compressed file
• The compressed proxysql binary
  See attached compressed file
• The compressed core dump (Note: if you're worried it may contain sensitive data, please
  See attached compressed file

Linux proxysql1 3.10.0-1062.12.1.el7.x86_64 #1 SMP Tue Feb 4 23:02:59 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux
[root@proxysql1 proxysql]# cat /etc/redhat-release
CentOS Linux release 7.7.1908 (Core)

[Tusamarco]

proxysql.zip

[Tusamarco]

I have also inserted the proxysql.db and the executable.
BTW please note that by mistake I had type proxysql --vrsion and instead raising an error and show the help it actully start proxysql.. u may want to catch that as well

[renecannao]

Hi Marco,

Thank you for the detailed report.
I am looking into fixing this ASAP (likely today).
Do you also have the proxysql.cnf file? I want to identify if something unusual was there.
Thank you

[renecannao]

@Tusamarco:

I was about to ask if you have a reproducible test case, because I couldn't understand how it was possible to trigger this crash.
Specifically, client_myds->myconn->variables[8]->value (character_set_connection) and all other character set variables are all set to NULL.
Established sessions shouldn't have NULL value for such variables, and I couldn't understand why this is the case in your core dump.
The answer is in the core dump itself:

(gdb) p client_myds->DSS 
$19 = STATE_SERVER_HANDSHAKE
(gdb) p client_myds->pkts_recv 
$20 = 0
(gdb) p client_myds->pkts_sent 
$21 = 1
(gdb) p client_myds->bytes_info 
$22 = {bytes_recv = 0, bytes_sent = 78}

In other words, the client wasn't authenticated yet, it didn't complete the handshake, thus the charset variables weren't initialized yet.
This means you managed to query stats_mysql_processlist while a client was in the middle of an authentication.
This is surely a bug, and we will fix it ASAP.

Thank you again for the report.

[Tusamarco]

@renecannao as u know use sysbench / sysbench-tpcc and the stresstool (java).
The issue I had was with the java app test. If you still need it I can pass to u the exact configuration file and command to run it.
Code is in github :
https://github.com/Tusamarco/stresstool

Anyhow I saw u have already merged with v2.013 will test and let u know

[Tusamarco]

@renecannao tested and seems resolved will keep you informed if any issue will raise again.
For me this can be close.