[sysbench] when use sysbench for arana test, not success

[dongzl]

What would you like to be added:

Why is this needed:

FATAL: `thread_run' function failed: /usr/share/sysbench/oltp_common.lua:432: SQL error, errno = 0, state = '00000':

sysbench oltp_read_write  --mysql-user=root --mysql-password=root --mysql-host=sharding-ip --mysql-port=3307 --mysql-db=s1_sbtest --histogram=on --report-interval=1 --time=300  --db-ps-mode=disable  --threads=300 --tables=250 --table_size=25000  --report-interval=1 --percentile=95 --skip-trx=on --mysql-ignore-errors=1062  --forced-shutdown=1 run

FATAL: `thread_run' function failed: /usr/share/sysbench/oltp_common.lua:432: SQL error, errno = 0, state = '00000': 
FATAL: mysql_stmt_execute() returned error 0 () for query 'SELECT c FROM sbtest97 WHERE id BETWEEN ? AND ?'

2023-02-28T09:58:06.810Z	ERROR	mysql/execute_handle.go:95	get dataset error 533: invalid sequence, expected 0 got 1 (errno 2013) (sqlstate HY000)
2023-02-28T09:58:06.810Z	ERROR	mysql/execute_handle.go:76	executor com_query error 580: invalid sequence, expected 1 got 2 (errno 2013) (sqlstate HY000)
2023-02-28T09:58:06.810Z	ERROR	mysql/execute_handle.go:76	executor com_query error 563: invalid sequence, expected 1 got 40 (errno 2013) (sqlstate HY000)

对应错误的sql信息FATAL: mysql_drv_query() returned error 2013 (invalid sequence, expected 1 got 40) for query 'SELECT c FROM sbtest188 WHERE id=12610'

[dobet]

yes, i use sysbench test. Before use arana test, I use Xiaomi/Gaea test, have this bug also.
And Gaea use Tidb parser also. So I think this is maybe tidb parse bug

[jjeffcaii]

It seems that a concurrency bug of packet ephemeral policy in the protocol layer. I'm trying my best to diagnose this problem.

[jjeffcaii]

BTW, the Xiaomi/Gaea and Arana, both used some codes from Vitess as the MySQL protocol implementation, maybe it's the reason that we have the same bug, I guess. 🤣

[dobet]

Yes， Xiaomi/Gaea have this bug also. vitess maybe have this bug， I don't test vitess， because it doesn't support auto read write split.

[jjeffcaii]

There's a backend connection leak problem, and it causes sending continuous requests (see the pic below).
It has been fixed in the #669, @dobet please check out this PR, thanks.