[
https://issues.apache.org/jira/browse/KYLIN-1434?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15200813#comment-15200813
]
Hao Chen commented on KYLIN-1434:
---------------------------------
[~liyang.g...@gmail.com] I thought clearly now, the simplest and backward
compatible solution is using coprocessor ResourceFilterObserver and don't need
to change the rowkey at all.
1) For old data where nothing needs to be changed and also fit for situation
where the kylin deployment is relatively small, we could only filter the data
in tomcat side:
{code}
hbase[scan(timeRangeFilter)] => tomcat[ResourceFilter(status/project/..)] =>
output
{code}
2) As the job history data increasing and the response become really slow,
user could enable the ResourceFilterObserver on the kylin_metadata table, then
we could enable the observer to filter the data in both hbase and tomcat side
and the response logic is same, while the difference is tomcat only need to
process much less filtered data:
{code}
hbase[scan(timeRangeFilter)+ResourceFilterObserver] =>
tomcat[ResourceFilter(status/project/..)] => same_output
{code}
The ResourceFilter interface:
{code}
/**
* Backward Compatible for both client side and remote side filter, especially
works for HBase on observer coprocessor
*/
public interface ResourceFilter extends Serializable {
/**
* @param resource RawResource
* @return false:throw resource, true: keep resource
*/
boolean filter(RawResource resource);
}
{code}
> Kylin Job Monitor API: /kylin/api/jobs is too slow
> --------------------------------------------------
>
> Key: KYLIN-1434
> URL: https://issues.apache.org/jira/browse/KYLIN-1434
> Project: Kylin
> Issue Type: Bug
> Affects Versions: v1.4.0
> Reporter: Hao Chen
>
> The API request for Job Monitor page like:
> {code}/kylin/api/jobs?limit=15&offset=15{code} takes more than 11 seconds for
> fetching only 15 row records (25.1 KB), which is too slow.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
