Billy, I don't see a need for "advanced user could enable it also." Do you
have such a case?

For the legacy cube, it can continue to build, the backend code won't be
removed during this phase.


2017-11-28 22:58 GMT+08:00 Billy Liu <[email protected]>:

> +1 to turn this feature off by default. The advanced user could enable it
> also.
>
> 2017-11-27 13:53 GMT+08:00 Luke Han <[email protected]>:
>
>> +1 to remove it from new release, people could backport to new version
>> with previous code
>>
>>
>> Best Regards!
>> ---------------------
>>
>> Luke Han
>>
>> On Sun, Nov 26, 2017 at 9:40 PM, ShaoFeng Shi <[email protected]>
>> wrote:
>>
>>> Last year I raised this discussion but didn't have follow-up action.
>>>
>>> Now we see there are still some new users misuse this feature, and then
>>> face performance and maintenance issues.
>>>
>>> In Kylin 2.1, the new "query pushdown" feature can forward the Cube
>>> unmatched queries to alternative query engines like Hive / SparkSQL. The
>>> raw data query is just such a scenario.
>>>
>>> So I think it is time to disable the RAW measure on Kylin now.  JIRA
>>> created for it: https://issues.apache.org/jira/browse/KYLIN-3062
>>>
>>> Please comment if you see any issue.
>>>
>>> 2016-12-19 22:03 GMT+08:00 Billy Liu <[email protected]>:
>>>
>>> > The experimental mode is system wide feature toggle. I think case by
>>> case
>>> > is more flexible. Most new features could have toggles, default are
>>> off.
>>> >
>>> > 2016-12-19 21:40 GMT+08:00 Luke Han <[email protected]>:
>>> >
>>> > > Beta or Experimental will also bring confusing for most of users.
>>> > >
>>> > > Maybe we could have something called "expert" or "experimental"
>>> model in
>>> > > system configuration.
>>> > >
>>> > > User will not see such content since they will be hidden by default
>>> but
>>> > > admin could set it to true if they confident to enable such features.
>>> > >
>>> > > How do you think?
>>> > >
>>> > >
>>> > > Best Regards!
>>> > > ---------------------
>>> > >
>>> > > Luke Han
>>> > >
>>> > > On Mon, Dec 19, 2016 at 11:24 AM, Xiaoyu Wang <[email protected]>
>>> > wrote:
>>> > >
>>> > > > I’m sorry for too long not maintain it!
>>> > > >
>>> > > > I agree with liyang to give a label to the RAW measure “Beta” or
>>> > others.
>>> > > >
>>> > > > I will improve it when I have time!
>>> > > >
>>> > > > 2016-12-19 10:21 GMT+08:00 Li Yang <[email protected]>:
>>> > > >
>>> > > > > Or display the Raw with "Beta" or "Experimental" to warn user
>>> that it
>>> > > is
>>> > > > > not a mature feature?
>>> > > > >
>>> > > > > On Fri, Dec 16, 2016 at 12:30 PM, 康凯森 <[email protected]> wrote:
>>> > > > >
>>> > > > > > +1.
>>> > > > > > But the "RAW" measure is still some useful, we could improve it
>>> > next
>>> > > > year
>>> > > > > > when we have time.
>>> > > > > >
>>> > > > > >
>>> > > > > > ------------------ 原始邮件 ------------------
>>> > > > > > 发件人: "ShaoFeng Shi";<[email protected]>;
>>> > > > > > 发送时间: 2016年12月15日(星期四) 中午12:05
>>> > > > > > 收件人: "dev"<[email protected]>;
>>> > > > > >
>>> > > > > > 主题: [Discuss] Disable/hide "RAW" measure in Kylin web GUI
>>> > > > > >
>>> > > > > >
>>> > > > > >
>>> > > > > > Hello developers and users,
>>> > > > > >
>>> > > > > > I have a proposal want to disucss here, which is about the
>>> "RAW"
>>> > > > measure
>>> > > > > in
>>> > > > > > Kylin.
>>> > > > > >
>>> > > > > > The RAW measure was developed to solve the requirement for
>>> getting
>>> > > raw
>>> > > > > data
>>> > > > > > when user drill down from high level to low levels. It's
>>> > performance
>>> > > > > would
>>> > > > > > be much better than fetching from source like Hive, so some
>>> users
>>> > > like
>>> > > > > it.
>>> > > > > > This blog introduces it:
>>> > > > > > https://kylin.apache.org/blog/2016/05/29/raw-measure-in-kylin/
>>> > > > > >
>>> > > > > > While it has some limitations:
>>> > > > > > 1) always use dictionary encoding, which means it couldn't
>>> support
>>> > > UHC;
>>> > > > > > since raw columns usually be transaction IDs, numbers, etc, the
>>> > > > building
>>> > > > > > cost is much higher than ordinary dimensions;
>>> > > > > > 2) the raw messages for a dimension combination are persisted
>>> in
>>> > one
>>> > > > big
>>> > > > > > cell,; When too many rows dumped in one cell, will get
>>> > BufferOverflow
>>> > > > > > error. This couldn't be predicted, so for an modeler or
>>> analyst he
>>> > > > > doesn't
>>> > > > > > know whether this feature could work when he creates the cube.
>>> > > > > > 3) seems no people maintains it;
>>> > > > > >
>>> > > > > > Based on above, I propose hiding this measure on Web GUI by
>>> > default,
>>> > > so
>>> > > > > > avoiding confusing user; If someone hear and want to use this,
>>> he
>>> > > still
>>> > > > > can
>>> > > > > > enable it by simply set in kylin.properties.
>>> > > > > >
>>> > > > > > Any input is welcomed.
>>> > > > > >
>>> > > > > > --
>>> > > > > > Best regards,
>>> > > > > >
>>> > > > > > Shaofeng Shi 史少锋
>>> > > > > >
>>> > > > >
>>> > > >
>>> > >
>>> >
>>>
>>>
>>>
>>> --
>>> Best regards,
>>>
>>> Shaofeng Shi 史少锋
>>>
>>
>>
>


-- 
Best regards,

Shaofeng Shi 史少锋

Reply via email to