Etsuro Fujita wrote:
> On 2015/09/02 16:40, Amit Langote wrote:
>> On 2015-09-02 PM 04:07, Albe Laurenz wrote:
>>> Amit Langote wrote:
>>>> On 2015-09-02 PM 03:25, Amit Kapila wrote:
>>>>> Will it handle deadlocks across different table partitions. Consider
>>>>> a case as below:
>>>>>
>>>>> T1
>>>>> 1. Updates row R1 of T1 on shard S1
>>>>> 2. Updates row R2 of T2 on shard S2
>>>>>
>>>>> T2
>>>>> 1. Updates row R2 of T2 on shard S2
>>>>> 2. Updates row R1 of T1 on shard S1
>>>
>>>> As long as shards are processed in the same order in different
>>>> transactions, ISTM, this issue should not arise? I can imagine it becoming
>>>> a concern if parallel shard processing enters the scene. Am I missing
>>>> something?
>>>
>>> That would only hold for a single query, right?
>>>
>>> If 1. and 2. in the above example come from different queries within one
>>> transaction, you cannot guarantee that shards are processed in the same order.
>>>
>>> So T1 and T2 could deadlock.
>
>> Sorry, I failed to see why that would be the case. Could you elaborate?
>
> I think Laurenz would assume that the updates 1. and 2. in the above
> transactions are performed *in a non-inherited manner*. If that's
> right, T1 and T2 could deadlock, but I think we assume here to run
> transactions over shards *in an inherited manner*.
Yes, but does every update affect all shards?
If I say "UPDATE t1 SET col = 1 WHERE id = 42" and the row with id 42
happens to be on shard S1, the update would only affect that shard, right?
Now if "UPDATE t2 SET col = 1 WHERE id = 42" would only take place on
shard S2, and two transactions issue both updates in different order,
one transaction would be waiting for a lock on shard S1, while the other
would be waiting for a lock on shard S2, right?
But maybe I'm missing something fundamental.
Yours,
Laurenz Albe
