Forward this to dev list also. Thanks.
Best Regards! --------------------- Luke Han ---------- Forwarded message ---------- From: Li Yang <[email protected]> Date: Thu, Oct 29, 2015 at 6:08 PM Subject: Re: Timestamp related issues To: [email protected] > 1. Is there any issue with Timestamp/Date values ? Timestamp testing is very limited on 1.x branch. All use cases I knew about uses date instead of timestamp. The 2.x branch has much better timestamp support. > 2. For measures with distinct count, it uses approximations with certain error rates, lowest of which is <1.22%. Does this guarantee that counts would be accurate ? The short answer is no 100% guarantee. The count distinct algorithm behind this is HyperLogLog [1]. Its error follows a normal distribution. The "< 1.22%" is brief of saying for 99.7% out of all the results, the error is <1.22% in theory. And there's still 0.3% results could go beyond the error. [1] https://en.wikipedia.org/wiki/HyperLogLog On Tue, Oct 27, 2015 at 12:45 PM, Chetan Dixit <[email protected]> wrote: > Hello Kylin Team, > > > > We are facing following issues while using Kylin could you please help. > > > > 1. Is there any issue with Timestamp/Date values ? > > We see issues in queries using “WHERE columnname = > timestamp ‘2015-07-23 10:30:00’ “ it does not return any results. > > If we use “WHERE columnname = ‘2015-07-23 10:30:00’ “ it > returns ERROR > > If use timestamp column in projection list, it truncates > the timestamp part i.e. 2015-07-23 10:30:00 to 2015-07-23 00:00:00 > > > > 2. For measures with distinct count, it uses approximations with > certain error rates, lowest of which is <1.22%. Does this guarantee that > counts would be accurate ? > > We have seen for a count of 1000 results as 982, 1000 etc. > > > > Thanks, > > Chetan > > >
