On 03.01.2011 18:49, Simon Riggs wrote:
> On Mon, 2011-01-03 at 18:35 +0200, Heikki Linnakangas wrote:
>> On 03.01.2011 18:29, Simon Riggs wrote:
>>> On Mon, 2011-01-03 at 18:08 +0200, Heikki Linnakangas wrote:
>>>
>>>> It works in read committed mode, because you acquire a new snapshot
>>>> after the LOCK TABLE, and anyone else who modified the table must commit
>>>> before the lock is granted. In serializable mode you get a serialization
>>>> error.
>>>
>>> If its not safe without this
>>>
>>> LOCK TABLE ... IN SHARE ROW EXCLUSIVE MODE
>>>
>>> then we should do that automatically, and document that.
>>
>> No we should not. The SQL standard doesn't require that, and it would
>> unnecessarily restrict concurrent updates on unrelated rows in the table.
>
> If we do that, then we definitely need a catch-all WHEN statement, so
> that we can say
>
> WHEN NOT MATCHED
> INSERT
> WHEN MATCHED
> UPDATE
> ELSE
> { INSERT into another table so we can try again in a minute
> or RAISE error }
>
> Otherwise we will silently drop rows. Throwing an error every time isn't
> useful behaviour.
An ELSE clause would be nice, but it's not related to the question at
hand. Only some serialization anomalities result in a row that matches
neither WHEN MATCHED nor WHEN NOT MATCHED. Others result in a duplicate
key exception, for example.
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com