From: Shawn Green
Date: August 1 2012 7:07pm
Subject: Re: How often the slave to pull data from master?
List-Archive: http://lists.mysql.com/mysql/227926
Message-Id: <50197E5E.4040906@oracle.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

On 8/1/2012 1:13 PM, Rick James wrote:
> That leads to another question...
> Does this 'separate cache' contain an extra copy (vs the 'only' copy) of the non-transacted events?
> I would expect that it must be an extra copy.
>

No. The non-transacted events are applied to the tables they belong to 
as they are executed.  We track them in this cache for replication 
purposes. Are you worried about a race between a crash and the time 
where the cache is applied to the main binary logs?

This is a very unlikely event for two reasons:

1) the non-transacted change must occur within a transaction

2) most users that use transactions only use transaction-enabled storage 
engines (InnoDB).

So in order to lose non-transacted changes from a crash that happens 
before the cache is appended to the binary log, you must
a) have a sequence of statements (or ROW-based changes) that is smaller 
than --binlog-cache-size limit
b) have a mixed-mode transaction.

If this is a valid concern for your usage pattern set 
--binlog-cache-size to a smaller value.
http://dev.mysql.com/doc/refman/5.5/en/replication-options-binary-log.html#sysvar_binlog_cache_size

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