percona 5.7.13 missing event in tpcc test with threadpool

[Jack Yu]

percona version: 5.7.13 commit e6fa0e9b586c483764bf791b696ef235dc9efaea

tpcc version: commit faa06df608d8bd36d540b0fc0042fd33debbde0c

1024 threads tpcc test (600 seconds) with threadpool, one transtion was sleeping for very long time (11770 seconds). All other transtions were blocked and the test can not finish. It seems like percona server missed a event from tpcc client. The probability of this bug occurrence is very low (about 1/15).

show processlist:

Id User Host db Command Time State Info Rows_sent Rows_examined
13074 root localhost tpcc1000 Sleep 11770 NULL 0 0
13115 root localhost tpcc1000 Execute 25 statistics SELECT d_next_o_id, d_tax FROM district WHERE d_id = ? AND d_w_id = ? FOR UPDATE 0 0
13241 root localhost tpcc1000 Execute 30 statistics SELECT d_next_o_id, d_tax FROM district WHERE d_id = ? AND d_w_id = ? FOR UPDATE 0 0
13296 root localhost tpcc1000 Execute 0 updating UPDATE warehouse SET w_ytd = w_ytd + ? WHERE w_id = ? 0 0
13439 root localhost tpcc1000 Execute 6 statistics SELECT d_next_o_id, d_tax FROM district WHERE d_id = ? AND d_w_id = ? FOR UPDATE 0 0
13564 root localhost tpcc1000 Execute 0 statistics SELECT d_next_o_id, d_tax FROM district WHERE d_id = ? AND d_w_id = ? FOR UPDATE 0 0
13614 root localhost tpcc1000 Execute 49 statistics SELECT d_next_o_id, d_tax FROM district WHERE d_id = ? AND d_w_id = ? FOR UPDATE 0 0
13617 root localhost tpcc1000 Execute 47 statistics SELECT d_next_o_id, d_tax FROM district WHERE d_id = ? AND d_w_id = ? FOR UPDATE 0 0
13618 root localhost tpcc1000 Execute 0 statistics SELECT d_next_o_id, d_tax FROM district WHERE d_id = ? AND d_w_id = ? FOR UPDATE 0 0
13636 root localhost tpcc1000 Execute 0 updating UPDATE warehouse SET w_ytd = w_ytd + ? WHERE w_id = ? 0 0
13920 root localhost tpcc1000 Execute 40 statistics SELECT d_next_o_id, d_tax FROM district WHERE d_id = ? AND d_w_id = ? FOR UPDATE 0 0
14012 root localhost tpcc1000 Execute 0 updating UPDATE district SET d_ytd = d_ytd + ? WHERE d_w_id = ? AND d_id = ? 0 0
14086 root localhost NULL Query 0 starting show processlist 0 0

percona server config file:
[mysqld]
log-error=error.log
explicit_defaults_for_timestamp = 1
secure-file-priv=NULL
skip-grant-tables
bind-address = 0.0.0.0
core-file
gdb

##### basic and group replication #####
innodb_flush_log_at_trx_commit=1
sync_binlog=1
server-id=1
port=13001
socket=/tmp/mysql.sock1
#log-bin=master.bin
binlog-format=row
gtid-mode=on
enforce-gtid-consistency
log-slave-updates
master-info-repository=TABLE
relay-log-info-repository=TABLE
transaction-write-set-extraction=MURMUR32
binlog-checksum=NONE

##### for performance purpose #####
innodb_file_per_table=0
innodb_buffer_pool_size=25G
innodb_checksum_algorithm=none
innodb_change_buffering=none
max_connections = 50000
table_open_cache = 8192
sort_buffer_size = 4M
join_buffer_size = 4M
query_cache_size = 0
query_cache_type = 0
innodb_doublewrite=0
innodb_flush_neighbors = 0
innodb_log_buffer_size = 256M
innodb_log_file_size = 5G
innodb_adaptive_hash_index=OFF
innodb_read_io_threads=2
innodb_write_io_threads=2
innodb_lru_scan_depth=8192
skip-ssl
innodb_use_native_aio=OFF
thread_handling=pool-of-threads
thread_pool_size=3

# for tpcc
max_prepared_stmt_count=262144
innodb_online_alter_log_max_size=256M

innodb_data_file_path=ibdata1:25G:autoextend
innodb_buffer_pool_instances = 4
innodb_page_cleaners = 1

tpcc test command:
./tpcc_start -S$SOCK -dtpcc1000 -uroot -w1000 -c1024 -r0 -l600 -f logs/tpcc_1024.log -t logs/tpcc_trx_1024.log

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-3499