Re: InnoDB AUTO-INC lock timeouts / deadlocks on inserts (using

2002-10-01 Thread iod iod 


Thank you for your response. 

We tried your example here on shell and it looks like InnoDB locks the table only for 
the first transaction after db startup and this lock is held for the whole transaction 
(like you observed).

But, contrary to this scenario we have experienced some strange behavior, such that 
table level AUTO-INC locks are being applied REPEATEDLY (duration of ENTIRE 
transaction), even though the database has undergone multiple, commited transactions 
previously. This behavior is very prominent when you have multiple processes with 
overlapping, concurrent transactions and you are executing DML statements in a loop (I 
tried a delay after each iteration to prevent a locks being held too long, but it did 
not seem to work). It so happens that the lock is not released by the process until 
the transaction is complete.

I am attaching a simplified perl script (concur-test.pl) and a shell script (go.sh) 
used to invoke these processes (URL: http://www.geocities.com/yoda_pk/innodbchk.tar.gz 
). Please create the table (SQL provided), run the shell script and observe the InnoDB 
monitor log for deadlocks. We have observed various AUTO-INC table level locks which 
are applied every time we run the tests (without restarting the mysql database engine) 
and these locks are held for the whole transaction and not just one particular 
statement, EVERY time. 

PLATFORM: RH Linux 7.3/MySQL-Max-3.23.52-1/Perl DBI

Looking forward to your observations.


regards,

IOD






Subject: Re: InnoDB AUTO-INC lock timeouts / deadlocks on inserts
(using
 Iod,
 
 - Original Message -
 From: iod iod [EMAIL PROTECTED]

 Newsgroups: mailing.database.mysql
 Sent: Thursday, September 26, 2002 8:14 AM
 Subject: InnoDB AUTO-INC lock timeouts / deadlocks on inserts
(using
 
 
 
  We have been using InnoDB with MySQL (MySQL-Max-3.23.52-1)
for transaction
 support in some complex situations in our application when
maintaining
 consistency is critical.
 
  Recently we did some stress testing by simulating 10
concurrent users
 (perl/DBD) doing large number of inserts on a few tables. Note
that these
 tables have indexed fields which are 'INT'  'AUTO_INCREMENT'.
 
  We are getting lock timeouts on concurrent inserts (in
concurrent
 TRANSACTIONS) very frequently, although it works fine for a
single
 non-concurrent transaction. Removing transactions eliminates the
deadlocks
 in concurrent runs, but we cannot guarantee the consistency of
our data.
 
  RELEVANT 'my.cnf' SETTINGS:
  innodb_lock_wait_timeout = 50
  innodb_thread_concurrency = 2 (to reduce thread contention)
 
  When we enabled InnoDB Monitor we saw messages like the
following:
 
  
 
  INSERT INTO BasketItems(BalQuantity, ...
  --- TRX HAS BEEN WAITING 46 SEC FOR THIS LOCK TO BE
GRANTED:
  TABLE LOCK table ca3/BasketItems trx id 0 26561327
lock_mode AUTO-INC
 waiting
 
  
 
  At times there are around 4 to 5 such simultaneous waits in
the log and
 ultimately the lock times out and the process terminates.
 
  After reading the docs, it seems like InnoDB applies a
table level lock on
 the table and so other inserts get blocked. But, considering the
time a
 process has to wait (  50s) it looks like the locks are
being held for a
 very long time. According to the InnoDB documentation
 (http://www.innodb.com/ibman.html#InnoDB_auto_inc):

 
  When accessing the auto-increment counter InnoDB uses
a special table
 level lock AUTO-INC lock which it keeps to the end of the
current SQL
 statement, not to the end of the transaction.
 
  Is it possible that the locks are being held for the WHOLE
transaction
 instead of just THAT SQL statement? Is it because InnoDB's
implementation of
 AUTO-INC locks needs to be more efficient?
 
 
 the auto-inc locks are necessary because MySQL binlogging
assumes that
 auto-inc values for a single SQL statement are determined by the
auto-inc
 value of the first insert in that statement.
 
 If you suspect that there is a bug and the auto-inc lock is held
longer, you
 should look at InnoDB Lock Monitor output.
 
 But also look at possible other bottlenecks. Log flushes, fsync,
etc.
 
 I tested this now. Note that after a table creation or a mysqld
startup, the
 very FIRST insert to the table acquires a next-key lock which is
held for
 the duration of the whole transaction. But subsequent inserts
only acquire
 the AUTO-INC lock, which is held only for the duration of the
SQL statement.
 
 
 User 1:
 
 
 heikki@hundin:~/mysql-max-3.23.52-pc-linux-gnu-i686/bin
mysql test
 Welcome to the MySQL monitor.  Commands end with ; or \g.
 Your MySQL connection id is 1 to server version: 3.23.52-max-log
 
 Type 'help;' or '\h' for help. Type '\c' to clear the buffer.
 
 mysql set autocommit = 0;
 Query OK, 0 rows affected (0.00 sec)
 
 mysql insert into kkk(b) values (100);
 Query OK, 1 row affected (0.00 sec)
 
 mysql commit;
 Query OK, 0 rows affected

Re: InnoDB AUTO-INC lock timeouts / deadlocks on inserts (using

2002-10-01 Thread iod iod 


Tuuri,

Thank you for the follow-up. It's great that our efforts were not in vain and I could 
contribute towards resolving the issue. I will be looking forward to the next release 
of InnoDB/MySQL-Max.

-
regards,

Afroze H. Zubairi (IOD)
-


- Original Message -
From: Heikki Tuuri [EMAIL PROTECTED]
Date: Tue, 1 Oct 2002 15:00:37 +0300
To: \iod iod\ [EMAIL PROTECTED]
Subject: Re: InnoDB AUTO-INC lock timeouts / deadlocks on inserts (using


 IOD,
 
 thank you for the bug report. I was able to repeat the problem with you
 scripts. The bug was that InnoDB failed to release at an SQL statement end
 such AUTO-INC locks which were only granted after a lock wait. They were
 only released at the transaction commit.
 
 Since AUTO-INC locks normally are held for a short time, AUTO-INC lock waits
 are rare, and many tests fail to notice this bug. Only under a stress test
 like the one you wrote this bug becomes visible.
 
 Below is a patch which fixes the bug.
 
 Best regards,
 
 Heikki
 Innobase Oy
 sql query
 
 ChangeSet
   1.1201 02/10/01 13:47:58 [EMAIL PROTECTED] +1 -0
   lock0lock.c:
 Fix bug: the AUTO-INC lock was held to the end of trx if it was granted
 after a lock wait
 
   innobase/lock/lock0lock.c
 1.16 02/10/01 13:47:39 [EMAIL PROTECTED] +13 -0
 Fix bug: the AUTO-INC lock was held to the end of trx if it was granted
 after a lock wait
 
 # This is a BitKeeper patch.  What follows are the unified diffs for the
 # set of deltas contained in the patch.  The rest of the patch, the part
 # that BitKeeper cares about, is below these diffs.
 # User: heikki
 # Host: hundin.mysql.fi
 # Root: /home/heikki/mysql
 
 --- 1.15/innobase/lock/lock0lock.c Sun Aug 11 18:34:57 2002
 +++ 1.16/innobase/lock/lock0lock.c Tue Oct  1 13:47:39 2002
 @@ -1863,6 +1863,19 @@
   ut_ad(mutex_own(kernel_mutex));
 
   lock_reset_lock_and_trx_wait(lock);
 +
 + if (lock_get_mode(lock) == LOCK_AUTO_INC) {
 +
 +  if (lock-trx-auto_inc_lock != NULL) {
 +  fprintf(stderr,
 + InnoDB: Error: trx already had an AUTO-INC lock!\n);
 +  }
 +
 +  /* Store pointer to lock to trx so that we know to
 +  release it at the end of the SQL statement */
 +
 +  lock-trx-auto_inc_lock = lock;
 + }
 
   if (lock_print_waits) {
   printf(Lock wait for trx %lu ends\n,
 
 - Original Message -
 From: iod iod [EMAIL PROTECTED]
 Newsgroups: mailing.database.mysql
 Sent: Tuesday, October 01, 2002 11:10 AM
 Subject: Re: InnoDB AUTO-INC lock timeouts / deadlocks on inserts (using
 
 
 
  Thank you for your response.
 
  We tried your example here on shell and it looks like InnoDB locks the
 table only for the first transaction after db startup and this lock is held
 for the whole transaction (like you observed).
 
  But, contrary to this scenario we have experienced some strange behavior,
 such that table level AUTO-INC locks are being applied REPEATEDLY (duration
 of ENTIRE transaction), even though the database has undergone multiple,
 commited transactions previously. This behavior is very prominent when you
 have multiple processes with overlapping, concurrent transactions and you
 are executing DML statements in a loop (I tried a delay after each iteration
 to prevent a locks being held too long, but it did not seem to work). It so
 happens that the lock is not released by the process until the transaction
 is complete.
 
  I am attaching a simplified perl script (concur-test.pl) and a shell
 script (go.sh) used to invoke these processes (URL:
 http://www.geocities.com/yoda_pk/innodbchk.tar.gz ). Please create the table
 (SQL provided), run the shell script and observe the InnoDB monitor log for
 deadlocks. We have observed various AUTO-INC table level locks which are
 applied every time we run the tests (without restarting the mysql database
 engine) and these locks are held for the whole transaction and not just one
 particular statement, EVERY time.
 
  PLATFORM: RH Linux 7.3/MySQL-Max-3.23.52-1/Perl DBI
 
  Looking forward to your observations.
 
  
  regards,
 
  IOD
  
 
 
 
 
 
  Subject: Re: InnoDB AUTO-INC lock timeouts / deadlocks on inserts
  (using
   Iod,
  
   - Original Message -
   From: iod iod [EMAIL PROTECTED]
 
   Newsgroups: mailing.database.mysql
   Sent: Thursday, September 26, 2002 8:14 AM
   Subject: InnoDB AUTO-INC lock timeouts / deadlocks on inserts
  (using
  
  
   
We have been using InnoDB with MySQL (MySQL-Max-3.23.52-1)
  for transaction
   support in some complex situations in our application when
  maintaining
   consistency is critical.
   
Recently we did some stress testing by simulating 10
  concurrent users
   (perl/DBD) doing large number of inserts on a few tables. Note
  that these
   tables have indexed fields which are 'INT'  'AUTO_INCREMENT'.
   
We are getting lock timeouts on concurrent inserts (in
  concurrent
   TRANSACTIONS) very