[Dovecot] Quota transaction bug?

[David Jonas]

Jasper Bryant-Greene wrote:
> On Tue, Jul 24, 2007 at 02:46:11PM -0700, David Jonas wrote:
>> Jasper Bryant-Greene wrote:
>>> On Tue, Jul 24, 2007 at 02:09:23PM -0700, David Jonas wrote:
>>>> < SELECT current FROM quota WHERE ...
>>>>> 513965019
>>>> < BEGIN
>>>> < INSERT INTO quota (current, ...
>>>> < COMMIT
>>>>> ??  (error_r appears to be (null))
>>> Unrelated to the original post, but the above would appear to be a bug to me.
>>>
>>> Because the SELECT is done before the transaction starts, the value in the
>>> INSERT which is based on the SELECT may no longer be consistent with the
>>> actual value in that table.
>> I don't think it uses the value. The INSERT statement that comes next
>> does an UPDATE if the key exists. The value represented as 0 here is the
>> change in quota value:
>>
>> INSERT INTO quota (current, path, username) VALUES (0, 'quota/storage',
>> 'user at example.com') ON DUPLICATE KEY UPDATE current = current + 0;
>>
>> So it's only added to or inserted, never changed then replaced.
> 
> OK, so absolutely nothing about the INSERT (including whether or not it is
> performed) depends on anything about the result from the SELECT?

I won't pretend to know for sure, but as far as I can tell, no. I never
poked around to find out why the SELECT is done. I'm guessing it was
done to see if the quota needed to be recalculated (negative values
cause a recalculation, according to the comments).

> Jasper