[
https://issues.apache.org/jira/browse/HIVE-21891?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Miklos Gergely updated HIVE-21891:
----------------------------------
Description:
DDLTask was a huge class, more than 5000 lines long. The related DDLWork was
also a huge class, which had a field for each DDL operation it supported. The
goal was to refactor these in order to have everything cut into more handleable
classes under the package org.apache.hadoop.hive.ql.exec.ddl:
* have a separate class for each operation
* have a package for each operation group (database ddl, table ddl, etc), so
the amount of classes under a package is more manageable
* make all the requests (DDLDesc subclasses) immutable - most of them are now
* DDLTask should be agnostic to the actual operations
* right now let's ignore the issue of having some operations handled by
DDLTask which are not actual DDL operations (lock, unlock, desc...)
In the interim time when there were two DDLTask and DDLWork classes in the code
base the new ones in the new package were called DDLTask2 and DDLWork2 thus
avoiding the usage of fully qualified class names where both the old and the
new classes were in use.
Step #12: rename DDLTask2 and DDLWork2, now that they are alone. Remove the old
DDLDesc. Instead of registering, now DDLTask finds the DDLOperations, and
registers them itself.
was:
DDLTask is a huge class, more than 5000 lines long. The related DDLWork is also
a huge class, which has a field for each DDL operation it supports. The goal is
to refactor these in order to have everything cut into more handleable classes
under the package org.apache.hadoop.hive.ql.exec.ddl:
* have a separate class for each operation
* have a package for each operation group (database ddl, table ddl, etc), so
the amount of classes under a package is more manageable
* make all the requests (DDLDesc subclasses) immutable
* DDLTask should be agnostic to the actual operations
* right now let's ignore the issue of having some operations handled by
DDLTask which are not actual DDL operations (lock, unlock, desc...)
In the interim time when there are two DDLTask and DDLWork classes in the code
base the new ones in the new package are called DDLTask2 and DDLWork2 thus
avoiding the usage of fully qualified class names where both the old and the
new classes are in use.
Step #11: extract the operations that left from the old DDLTask, and move them
under the new packages.
> Break up DDLTask - cleanup
> --------------------------
>
> Key: HIVE-21891
> URL: https://issues.apache.org/jira/browse/HIVE-21891
> Project: Hive
> Issue Type: Sub-task
> Components: Hive
> Affects Versions: 3.1.1
> Reporter: Miklos Gergely
> Assignee: Miklos Gergely
> Priority: Major
> Labels: refactor-ddl
> Fix For: 4.0.0
>
>
> DDLTask was a huge class, more than 5000 lines long. The related DDLWork was
> also a huge class, which had a field for each DDL operation it supported. The
> goal was to refactor these in order to have everything cut into more
> handleable classes under the package org.apache.hadoop.hive.ql.exec.ddl:
> * have a separate class for each operation
> * have a package for each operation group (database ddl, table ddl, etc), so
> the amount of classes under a package is more manageable
> * make all the requests (DDLDesc subclasses) immutable - most of them are now
> * DDLTask should be agnostic to the actual operations
> * right now let's ignore the issue of having some operations handled by
> DDLTask which are not actual DDL operations (lock, unlock, desc...)
> In the interim time when there were two DDLTask and DDLWork classes in the
> code base the new ones in the new package were called DDLTask2 and DDLWork2
> thus avoiding the usage of fully qualified class names where both the old and
> the new classes were in use.
> Step #12: rename DDLTask2 and DDLWork2, now that they are alone. Remove the
> old DDLDesc. Instead of registering, now DDLTask finds the DDLOperations, and
> registers them itself.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
