Thank you!
I just tested the changes with queries like:
SELECT dt, SUM(metric) FROM table WHERE dt IN ('2015-10-01', '2015-10-02')
GROUP BY dt;
SELECT dt, SUM(metric) FROM table WHERE dt BETWEEN '2015-10-01' AND
'2015-10-07' GROUP BY dt;
before every partition was scanned, after the changes only relevant partitions
are scanned.
Very useful performance change, thanks again.
On October 28, 2015 at 4:02:19 AM, Li Yang ([email protected]) wrote:
Em.. I searched commits on KYLIN-747 and get nothing too. Thought it was
covered by fix to some other JIRA.
Anyway, I cooked some test cases to query non-existing values, and made
some further optimization. Now the ever-false scan range can be pruned.
https://github.com/apache/incubator-kylin/commit/f96600e89a5a2c0e533cea86ad6a73b9451bcddc
On Tue, Oct 27, 2015 at 12:03 PM, Vadim Semenov <[email protected]> wrote:
> Hi everyone,
>
> I'm trying to find the commit where this issue was fixed (
> https://issues.apache.org/jira/browse/KYLIN-747).
> Could you point me?
>
> We have a cube that is partitioned by date, and when we query using SELECT
> * FROM table WHERE dt = '2015-10-01', I see in the logs:
> "Can't translate value 2015-10-01 to dictionary ID, roundingFlag 0. Using
> default value \xFF",
> which translates into a huge scan.
