Am 02.03.2015 um 15:22 schrieb Jean Weisbuch:
Just as a side note, its possible to remove the TRANSACTIONAL=x option by adding TRANSACTIONAL=default on the ALTER statement.
wouldn't a "ALTER TABLE `systemevents` ENGINE=InnoDB DEFAULTS;" be more logical which should be default in case of strict mode or at least give a clear message about it?
A similar issue arises if for example ROW_FORMAT=PAGE is used on the ARIA table as it doesnt exists on InnoDB, it will issue a warning "InnoDB: assuming ROW_FORMAT=COMPACT." but the resulting table will show ROW_FORMAT=PAGE even if it uses the COMPACT row format...
and that is really confusing and may lead to weird problems it should not assume but set it to ROW_FORMAT=COMPACT and only isse errors if that for whatever reasons fails what we now have is that you can't trust the output of params at all
Le 02/03/2015 15:18, Reindl Harald a écrit :
Am 02.03.2015 um 15:03 schrieb Ian Gilfillan:
On 02/03/2015 15:37, Reindl Harald wrote:
Am 02.03.2015 um 14:26 schrieb Ian Gilfillan:
This is probably because strict mode is set. If unset, it should return a warning, not an error. From MariaDB 10.1.2, you can use: SET STATEMENT sql_mode='' FOR ALTER TABLE `systemevents` ENGINE = InnoDB; which will unset strict mode just for that query, otherwise for older versions just set it off and on again for the session to alter.
indeed *but* why does "TRANSACTIONAL=0" don't go away after convert to InnoDB?
Why is so that the change is reversible
that makes no sense, keep options which are invalid for the target engine is completly wrong and leads to the above problems that you can't convert tables only in the dirty default mode
if you now convert to InnoDB "TRANSACTIONAL=0" makes no sense if you later decide to convert back to Aria you need to set that flag in a follow-up step - but keep invalid and dirty options is just asking for troubles until "alter table" has no way to delete them
Obviously this may not be desirable in many cases. See https://mariadb.atlassian.net/browse/MDEV-5867 for a similar discussion and a fix that was applied in 10.0.13
it is not desirebale in *any* sane case