Sergey Shelukhin created HIVE-9790:
--------------------------------------
Summary: Hybrid Hybrid Grace Hash Join: improve side file
serialization
Key: HIVE-9790
URL: https://issues.apache.org/jira/browse/HIVE-9790
Project: Hive
Issue Type: Improvement
Reporter: Sergey Shelukhin
We have discussed it in the past; the current method is very wasteful, if
serializes helper object for each row (so there's expensive serialization and
also bunch of unneeded stuff serialized), whereas at the "memory-insert vs
side-file-spill" decision point we can produce bytes that are directly usable
by hashtable in one method call. So we should do that... at load point, again
there's no expensive deserialization, and no helpers, bytes can go into
hashtable directly pretty much
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
