Re: bzr commit into mysql-next-mr-rpl-merge branch (alfranio.correia:3202) WL#2775

[Alfranio Correia]

Hi all,

Please, find a new patch at http://lists.mysql.com/commits/121790.
I already pushed the patch to the mysql-next-mr.crash-safe.

If there is anything left I will commit diffs.

Cheers.


On 10/20/2010 11:05 AM, He Zhenxing wrote:
> Luís Soares wrote
>> On 10/20/2010 10:12 AM, Alfranio Correia wrote:
>>> Hi Jasonh, Luis,
>>>
>>> This is the only thing left to be fixed.
>>> As you know I will be off-line today but can you (Luis and Jasonh)
>>> discuss and tell
>>> me what to do? If not possible, let's have a call tomorrow.
>>
>> Lets have a quick call tomorrow then.

I could not come up with a solution for this.

If we try to execute an alter table when the server is starting up, we get a segmentation
fault.
This happens because several structures/objects are not ready at this point.
And, the alter table code is too complex to extract the necessary low-level functions
without
the risky of missing anything.

So I suggest to do what follows:

   PROPOSAL 1:
   . Let's keep the code as it and file a bug report that I will be addressed as a
post-push fix.
   . In this bug report we can try three different solutions:
     1. Call low-level functions to change the type of the engine thus avoiding using the
       parser.
     2. Figure out if the necessary objects can be created and use the current idea.
     3. Move the call to another point in the code where the problem does not happen.

   PROPOSAL 2:
   . Try to fix the problem by using either (1), (2) or (3).

I cannot promise an ETA and will need help in both proposals.

Cheers.

>>
>
> OK
>
>>>
>>> Cheers.
>>>
>>>
>>> On 10/19/2010 06:14 AM, He Zhenxing wrote:
>>>> Alfranio Correia wrote:
>>>>>> R6. Do we allow changing repo table engine dynamically? If not
>>>>>>        then I think Rpl_info_table::do_is_transactional() no
> need
>>>>>>        to check the table every time, I think this can be
> checked
>>>>>>        and saved the result when initializing the info table.
>>>>>
>>>>> Yes. It is possible to change it dynamically. A super user can
>>>>> always execute "alter table".
>>>>>
>>>>
>>>> I tent to think this is not a good idea to allow repo table engine to be
>>>> changed at runtime, it might cause unexpected behavior,  so I'd strongly
>>>> suggest to disable this if you do not have any good use cases for this
>>>> feature.
>>>>
>>>
>>
>>
>
>
>


Attachment: [text/x-patch] alter-table.diff