Thanks ShaoFeng, I have copy a rawv2 measure without dictionary instead, all memory metrics have been around normal level since then.
> On Jul 31, 2019, at 1:16 PM, ShaoFeng Shi <[email protected]> wrote: > > This is a limitation of the raw measure; It uses a dictionary to compress the > values, but that may make the dictionary very big. This is one of the reasons > that Kylin mark the raw measure as deprecated. > > Best regards, > > Shaofeng Shi 史少锋 > Apache Kylin PMC > Email: [email protected] <mailto:[email protected]> > > Apache Kylin FAQ: https://kylin.apache.org/docs/gettingstarted/faq.html > <https://kylin.apache.org/docs/gettingstarted/faq.html> > Join Kylin user mail group: [email protected] > <mailto:[email protected]> > Join Kylin dev mail group: [email protected] > <mailto:[email protected]> > > > > > you Zhuang <[email protected] <mailto:[email protected]>> > 于2019年7月15日周一 下午11:27写道： > I use raw measure to serve original records consults. But why does raw > measure need dictionary? If I have a lot of raw measures, performance is > slow. I see code below : > > @Override > public List<TblColRef> getColumnsNeedDictionary(FunctionDesc functionDesc) { > TblColRef literalCol = functionDesc.getParameter().getColRefs().get(0); > return Collections.singletonList(literalCol); > } > > My raw measures are always double, not big. So I don’t need dictionary at all. > > I use raw measures for historical reasons, so I expect someone to tell me why > and how I can modify to use raw bytes.
