Hi
Libing,
Alfranio Correia wrote:
> Hi Libing,
>
> Thank you for the review.
>
>
> Libing Song wrote:
>> Hi Alfranio,
>>
>> Nice work, but a problem need to be clarified.
>>
>> The option --ignore-causality introduces an another problem in SBM.
>> The following statements will cause a inconsistency between master and
>> slave.
>>
>> CREATE TABLE t1(c1 INTEGER) ENGINE=Innodb;
>> CREATE TABLE t2(c1 INTEGER) ENGINE=MyISAM;
>> BEGIN ;
>> INSERT INTO t1 VALUES (1),(2);
>> INSERT INTO t2 SELECT * FROM t1;
>> COMMIT;
>>
>> so, I think that locking the N_table before starting a transaction is
>> better.
>
> I don't think we should do this but if you want to discuss this option
> I can elaborate my opinion against this idea.
>
>> Maybe It is difficult to implement, however give users a advice is easy.
>
> I agree with you. I think we should print a warning message in such case.
>
I thought about this while changing the patch and think we should only document
this and not print any warning message.
In general, it is unsafe to mix non-transactional engines and transactional engines in 5.1
and we do not print any messages. So, I don't think we should start printing messages now.
We already changed the rep+3 to correctly print warning messages.
>> I propose --ignore-causality is a session variable which can be changed
>> at runtime,
>> if it has to be used.
>> this will reduce the option's influence.
Done.
>
> I agree with you.
>
> Cheers.
>
