1. Creating a cube right from a SQL is charming. Had discussed this idea, but never come to implement. A big issue is user still have to learn cube concepts, otherwise the auto-generated cube is either too big or the query is too slow. Still this is a good idea to bootstrap new users.
2. Currently Kylin throws exception when no matching cube. You can catch the exception in the frontend and guide user to hive query GUI. That's perfectly doable, and I believe, very useful too. You may start by calling Kylin Restful API from javascript [1], and detect error response that is caused by no matching cube. [1] http://kylin.incubator.apache.org/docs/howto/howto_use_restapi_in_js.html On Mon, Sep 14, 2015 at 4:50 PM, Leon Zhang <[email protected]> wrote: > Thank you Li Yang for you quick reply. > > 1. From a newbie's view, create a cube is too complex. How about if user > provide a SQL, kylin can automatically generate the model json and cube > json? > > 2. IMHO, SparkSQL can be faster, but, it may still take hours. Can kylin > provide such API, that *test* if the query can be handled by any cube. If > not, user can *directly* call hive-server2 for slow query? > > I am willing to hear about detailed advises, so that I can hack Kylin to > implement these potential features. > > Thanks. > > > On Mon, Sep 14, 2015 at 4:33 PM, Li Yang <[email protected]> wrote: > > > 1. You can create a view using the SQL, then let the view be Kylin fact > > table and build cube from it. > > > > 2. Technically yes, but the routing feature is disabled long ago. We had > > the same re-route idea in early PoC of Kylin. Later it turns out hive is > > too slow, and when mixed up, a query sometimes returns in seconds, > > sometimes in hours, the user experience is very bad. The feature is then > > disabled and never maintained any more. The idea is still valid, we may > > implement when hive speed catches up, or route to some faster SQL engine, > > like SparkSQL. KYLIN-742 < > https://issues.apache.org/jira/browse/KYLIN-742> > > > > > > On Mon, Sep 14, 2015 at 2:11 PM, Leon Zhang <[email protected]> wrote: > > > > > Hi, Kylin Developers, > > > > > > I am a newbie to Kylin system. During I investigate this awesome > > system, > > > these questions come to me: > > > > > > 1. Can I build a cube from a SQL? For learn_kylin example, the SQL for > > > "kylin_cube_sales" is: > > > > > > ``` sql > > > SELECT > > > FACT_TABLE.PART_DT > > > ,FACT_TABLE.LEAF_CATEG_ID > > > ,FACT_TABLE.LSTG_SITE_ID > > > ,LOOKUP_2.META_CATEG_NAME > > > ,LOOKUP_2.CATEG_LVL2_NAME > > > ,LOOKUP_2.CATEG_LVL3_NAME > > > ,FACT_TABLE.LSTG_FORMAT_NAME > > > ,FACT_TABLE.PRICE > > > ,FACT_TABLE.SELLER_ID > > > FROM DEFAULT.KYLIN_SALES as FACT_TABLE > > > INNER JOIN DEFAULT.KYLIN_CAL_DT as LOOKUP_1 > > > ON FACT_TABLE.PART_DT = LOOKUP_1.CAL_DT > > > INNER JOIN DEFAULT.KYLIN_CATEGORY_GROUPINGS as LOOKUP_2 > > > ON FACT_TABLE.LEAF_CATEG_ID = LOOKUP_2.LEAF_CATEG_ID AND > > > FACT_TABLE.LSTG_SITE_ID = LOOKUP_2.SITE_ID > > > ``` > > > > > > 2. Kylin is super fast, can I route *unmatched* query to slow hive > > engine? > > > For example, the simple query like "select * from kylin_sales" return a > > > empty result set. Can I route query like this to the hive engine? > > > > > > > > > Thanks. > > > > > >
