Hello Mogens,

On 8/18/2018 2:32 PM, Mogens Melander wrote:

I think I remember this from way back.

You could ask for a lock, and get an OK if it is safe.

Something like, if there is pending transactions, on your target tables,
you would get a NO.

But then again. I could be wrong, and Shawn is the authority on this.

Your request for a lock would have waited until all existing readers or writers (depending on the type of lock you asked for) had finished using the tables you wanted to lock. By extension, that means that any transactions active against the tables you wanted to lock would have also needed to have committed or rolled back before your request would have been granted. Any new actions against the table would have been queued up behind your LOCK request. This has confused more than one DBA as they didn't realize that the LOCK was going to be such a tight bottleneck.

These kinds of whole table locks live above the blocking/locking coordination of the individual storage engines or the transaction control code. They are managed in the "server layer" of our code.

This separation of scope is one reason why blending transactional and non-transactional tables in the same data management process is generally frowned on. Either be all-transactional (InnoDB) or not. The behavior will be easier to predict allowing your developers to use either the transaction control commands (BEGIN/COMMIT/ROLLBACK/... ) or the LOCK commands with confidence.

Shawn Green
MySQL Senior Principal Technical Support Engineer
Oracle USA, Inc. - Integrated Cloud Applications & Platform Services
Office: Blountville, TN

Become certified in MySQL! Visit https://www.mysql.com/certification/ for details.

=== original thread ===

On 2018-08-18 23:59, shawn l.green wrote:
Hello Jeff,

On 8/13/2018 12:05 PM, j...@lxvi.net wrote:
Hello, I have read through several pages of the reference manual, and
I've seen several instances where it is stated that LOCK TABLES (and
UNLOCK TABLES) is not allowed in a stored procedure, but so far, I
haven't found an explanation as to *why* that is. Could someone please
enlighten me?


Normally, the list is more responsive than this. This is a pretty easy
question and someone usually handles those before I need to step in as
a backstop.

The key why you cannot execute a LOCK TABLE command within a stored
program is here:
LOCK TABLES is not transaction-safe and implicitly commits any active
transaction before attempting to lock the tables.

Stored programs execute under the scope of the transaction in which
they are started. That determines which sets of rows are "visible" to
the routine and sets boundaries on what may be committed or rolled
back should the need arise.

(a simple example)
* your session: START TRANSACTION
* your session: ...other data activity ...
* your session (INSERT ... )
  * causes an INSERT trigger to fire
    * which calls a stored procedure

If that stored procedure or that trigger called a LOCK TABLE command,
it would forcibly COMMIT the existing transaction you had been working
within until that moment.  Your half-completed work would have become
fully committed even if a later step had needed you to issue a
ROLLBACK command.

Note, even if you are not in a multi-statement transaction that any
stored programs called by or executed within the scope of your user
command are part of that little mini (auto-committed) transaction.

Does that help?

Shawn Green
MySQL Senior Principal Technical Support Engineer
Oracle USA, Inc. - Hardware and Software, Engineered to Work Together.
Office: Blountville, TN

MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:    http://lists.mysql.com/mysql

Reply via email to