Re: bzr commit into mysql-5.1 branch (hezx:2632) Bug#37051

[He Zhenxing]

Sven Sandberg wrote:
> Ingo Strüwing wrote:
> >> He Zhenxing wrote:
> >>> #At file:///media/sda3/work/mysql/bzrwork/b37051/5.1-rpl-new/
> >>>
> >>>  2632 He Zhenxing    2008-07-25
> >>>       BUG#37051 Replication rules not evaluated correctly
> > ...
> >> Please begin each test case with comments describing:
> > ...
> >>  - A list of related bugs (BUG#37051).
> > 
> > And please don't forget the synopsis/title.
> 
> I agree.
> 

good.

> >> Please clean up (drop all tables and sync_slave_with_master) at the end
> >> of the test.
> > 
> > Isn't there an include file that should be used to clean up replication
> > tests?
> 
> Yes, include/master-slave-end.inc should be sourced, so that the slave 
> is stopped. Also, all tables and other database objects must be dropped, 
> and all variables must be reset to their old values.
> 

good, thanks

> >>> +      user_var_entry *var_entry= (user_var_entry
> >>> *)hash_search(&thd->user_vars,
> >>> +                                                               (const
> >>> uchar *)name.str,
> >>> +                                                              
> >>> name.length);
> >>> +      if (var_entry)
> >>> +        table_map_for_update=
> >>> (table_map)var_entry->val_int(&null_value);
> > 
> > After reading through the patch once, I wondered if we don't need to
> > distinguish between old masters, where we still need to analyze the
> > tables in the slave, and new masters, that send the map of update tables.
> >
> > Do I understand the patch correctly in so far that the assignment above
> > simply overrides the analyze done by the server, if the master sent a
> > table map?
> 
> As I understand it, the old code for checking replication rules is in 
> sql_parse.cc just below the patch. This code is guarded by 
> lex->sql_command!=SQLCOM_MULTI_UPDATE. Hence, before the patch, 
> multi-table UPDATE statements were simply not checked for replication 
> rules. The new case introduced by the patch is guarded by 
> lex->sql_command==SQLCOM_MULTI_UPDATE. So no query executes both checks.
> 

For new masters, both the new and the old checks are processed if the
filter rules does not ignore the statement.

> > If yes, would there be savings if we skip the analyze in case we have
> > the map from the master?
> > 
> > ...
> > 
> > Regards
> > Ingo
> 
> -- 
> Sven Sandberg, Software Engineer
> MySQL AB, www.mysql.com
>