Hi Anil,

Group by queries with collocate = true will be correct if:
for every group there is only partition that contains all data of this
group.
Ignite does not know if group data is collocated for a query, so with this
option we can give a hint to Ignite.

For example: In this case, we can make grouping and apply Having condition on
map stage, that
results in reducing network traffic.

For non-collocated groups we can't apply Having on map stage as we have
only part of data at this time.
So, in this case with collocate = true we can get wrong results.



On Tue, Feb 7, 2017 at 6:29 AM, Anil <[email protected]> wrote:

> Hi,
>
> May I know what do you mean by "When used incorrectly" ?
>
> group by queries with collocate = true is not incorrect usage or i
> misunderstood this ?
>
> Thanks.
>
> On 7 February 2017 at 01:26, vkulichenko <[email protected]>
> wrote:
>
>> Well, I would agree that the flag is confusing, but your understanding is
>> not
>> correct. When used incorrectly, this flag does breaks query result because
>> it forces Ignite to change execution plan to more optimal. This
>> optimization
>> doesn't always work though (that's actually the reason why it's
>> configurable).
>>
>> -Val
>>
>>
>>
>> --
>> View this message in context: http://apache-ignite-users.705
>> 18.x6.nabble.com/NOT-IN-in-ignite-tp9861p10458.html
>> Sent from the Apache Ignite Users mailing list archive at Nabble.com.
>>
>
>


-- 
Best regards,
Andrey V. Mashenkov

Reply via email to