USER_STATISTICS shows incorrect value for number of concurrent connections

[Swany]

mysql> select * from USER_STATISTICS where user='cactiuser'\G
*************************** 1. row ***************************
                  USER: cactiuser
     TOTAL_CONNECTIONS: 1002
CONCURRENT_CONNECTIONS: 0
        CONNECTED_TIME: 23918
             BUSY_TIME: 41
              CPU_TIME: 32
        BYTES_RECEIVED: 1823054
            BYTES_SENT: 933888
  BINLOG_BYTES_WRITTEN: 0
          ROWS_FETCHED: 698035
          ROWS_UPDATED: 4422
       TABLE_ROWS_READ: 19428245
       SELECT_COMMANDS: 7832
       UPDATE_COMMANDS: 2398
        OTHER_COMMANDS: 4620
   COMMIT_TRANSACTIONS: 9306
 ROLLBACK_TRANSACTIONS: 0
    DENIED_CONNECTIONS: 0
      LOST_CONNECTIONS: 0
         ACCESS_DENIED: 0
         EMPTY_QUERIES: 2625
1 row in set (0.00 sec)

Note that CONCURRENT_CONNECTIONS shows 0, but it should always be at least 1, and in this case it should be much higher:

In the following snippet from SHOW PROCESSLIST you can see multiple concurrent connections by this user:
*************************** 19. row ***************************
     Id: 389
   User: cactiuser
   Host: localhost
     db: cacti
Command: Sleep
   Time: 710
  State:
   Info: NULL
*************************** 20. row ***************************
     Id: 390
   User: cactiuser
   Host: localhost
     db: cacti
Command: Sleep
   Time: 3937
  State:
   Info: NULL
*************************** 21. row ***************************
     Id: 391
   User: cactiuser
   Host: localhost
     db: cacti
Command: Sleep
   Time: 710
  State:
   Info: NULL
21 rows in set (0.00 sec)

[Swany (greenlion)]

Percona Server 5.5.8-55

[Valentine Gostev (longbow)]

Oleg, please examine the patch source (I could not find the condition when concurrent_connections is different from zero)

[Oleg Tsarev (tsarev)]

I found following:
userstats.patch was incorrectly ported from google-patches.

[Oleg Tsarev (tsarev)]

In fix I did following (comment from commit):

------
Add set_connections function (related to incomplete port from Google patch) and add call to send_user_stats and send_thread_stats (bug #728082, bug #608027). Without this call statistcs from separated thread didn't aggregated to global_[client|thread|user]_stats.
------

Every thread contains STATS diff_stats structure, which collected information about CPU_Time/other counters after every executed query.
Function "set_connection_stats" extract this information from thread to global_[client|thread|user]_stats.

Port from Google patch was incomplete - set_connection_stats missed, and information doesn't moved from THD to HASH table.

[Oleg Tsarev (tsarev)]

Additional information. Concurrent_connection is update in increment_count_by_name ([client|user]_stats) and increment_count_by_id (thread_stats).

Before fix this information updated just after thread complete (and just for this thread), as result Concurrent_connection always zero.

After my fix, set_connection_stats clean all Concurrent_connections counter, and calculate it directly from every thread (set_connection_stats called on SHOW [CLIENT|THREAD|USER]_STATS).

[Laurynas Biveinis (laurynas-biveinis)]

https://github.com/percona/percona-server/pull/1135, https://github.com/percona/percona-server/pull/1178, https://github.com/percona/percona-server/pull/1179

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-1193