[ https://issues.apache.org/jira/browse/HIVE-12050?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Sergey Shelukhin updated HIVE-12050: ------------------------------------ Description: With directory listing, ETL vs BI decision, local cache, metastore cache, with PPD, file footers, and combination thereof (e.g. most splits are processed via metastore PPD but some files are not cached and we need to make ETL vs BI decision), some of which are blocking and some not, -I want to write ORC split generation in Erlang- strategies are no longer the best model to schedule all the work. Some messaging or task queue based model might be better where each work item that is blocking (dir listing, file read, metastore call, etc.) generates list of things, and things are further processed until all things are splits and there are no more other things to process. (was: With directory listing, ETL vs BI decision, local cache, metastore cache, with PPD, file footers, and combination thereof (e.g. most splits are processed via metastore PPD but some files are not cached and we need to make ETL vs BI decision), some of which are blocking and some not, -I want to write ORC split generation in Erlang- strategies are no longer the best model to schedule all the work. Some messaging or task queue based model might be better where e.g. directory listers generate file lists that are processed by decision making strategy that might go to metastore or cache or files, and remaining files might again be processed thru decision making or thru a different cache or files, etc.) > change ORC split generation to use a different model > ---------------------------------------------------- > > Key: HIVE-12050 > URL: https://issues.apache.org/jira/browse/HIVE-12050 > Project: Hive > Issue Type: Bug > Reporter: Sergey Shelukhin > > With directory listing, ETL vs BI decision, local cache, metastore cache, > with PPD, file footers, and combination thereof (e.g. most splits are > processed via metastore PPD but some files are not cached and we need to make > ETL vs BI decision), some of which are blocking and some not, -I want to > write ORC split generation in Erlang- strategies are no longer the best model > to schedule all the work. Some messaging or task queue based model might be > better where each work item that is blocking (dir listing, file read, > metastore call, etc.) generates list of things, and things are further > processed until all things are splits and there are no more other things to > process. -- This message was sent by Atlassian JIRA (v6.3.4#6332)