[ 
https://issues.apache.org/jira/browse/KYLIN-4579?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Zhong Yanghong updated KYLIN-4579:
----------------------------------
    Description: 
If there're many backend kylin instances, kylin instances will broadcast 
metadata change asynchronously. In this case, when user try to create a cube by 
program, whose normal process is to
# load tables
# create model
# create cube

Since this process is not atomic and the related three requests will be sent to 
backend instances randomly, it probably causes exceptions due to the metadata 
update delay.

For example, if there're two kylin instances, K1 & K2. User sends "load tables" 
request to K1, then sends "create model" request to K2. If K2 is not notified 
with metadata update in time, it will fail to initialize the model and throw 
exceptions.

To deal with this issue, it's better to create a fat API to create cube & 
delete cube, which will make it possible to send all the step request to one 
server.

> Add fat API to create cube & delete cube
> ----------------------------------------
>
>                 Key: KYLIN-4579
>                 URL: https://issues.apache.org/jira/browse/KYLIN-4579
>             Project: Kylin
>          Issue Type: Improvement
>          Components: REST Service
>            Reporter: Zhong Yanghong
>            Assignee: Zhong Yanghong
>            Priority: Major
>
> If there're many backend kylin instances, kylin instances will broadcast 
> metadata change asynchronously. In this case, when user try to create a cube 
> by program, whose normal process is to
> # load tables
> # create model
> # create cube
> Since this process is not atomic and the related three requests will be sent 
> to backend instances randomly, it probably causes exceptions due to the 
> metadata update delay.
> For example, if there're two kylin instances, K1 & K2. User sends "load 
> tables" request to K1, then sends "create model" request to K2. If K2 is not 
> notified with metadata update in time, it will fail to initialize the model 
> and throw exceptions.
> To deal with this issue, it's better to create a fat API to create cube & 
> delete cube, which will make it possible to send all the step request to one 
> server.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to