[
https://issues.apache.org/jira/browse/TAJO-744?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13994817#comment-13994817
]
Hyunsik Choi commented on TAJO-744:
-----------------------------------
Hi [~alvinhenrick],
I leaved some comment the reviewboard. Today, I'd like to continue the
discussion. Earlier, I only suggested the parts that you can replace with
existing utility methods. At that time, I said that I'll leave comments about
{{PARTITION_METHODS_STORE}} schema. I'm sorry for late response.
Today, I'm going to discuss the partition entry schema. I have thought your
proposed schema for a few of days. The schema seems to work in all cases
efficiently with indexing techniques on right data types.
My one concern is that the table will be very sparse. For dense table, we can
use several TEXT columns instead of various data type columns in order to store
all partition columns. Actually, I have no strong idea that which one is better
way. I'd like to hear your thought.
Warm Regards,
Hyunsik
> ALTER TABLE ADD/DROP PARTITION statement
> ----------------------------------------
>
> Key: TAJO-744
> URL: https://issues.apache.org/jira/browse/TAJO-744
> Project: Tajo
> Issue Type: New Feature
> Components: catalog
> Affects Versions: 0.9.0
> Reporter: Hyunsik Choi
> Assignee: Alvin Henrick
> Fix For: 0.9.0
>
> Attachments: TAJO-744.Henrick-140423.01.patch.txt
>
>
> Currently, Tajo does not manage partitioned directly. In Tajo, each partition
> is just a directory. For each query, a logical planner traverses matched
> directories in HDFS according to partition predicates.
> This approach is not efficient especially in the environment where the number
> of partitions are very large. It also makes partition management hard.
> Tajo should manage partitions directly by using ALTER TABLE ADD/DROP
> PARTITION statements. A number of partition entries should be stored in the
> underlying database that catalog uses.
> {code:title=Synopsis of ALTER TABLE ADD/DROP PARTITION}
> ALTER TABLE table_name [IF NOT EXISTS] ADD COLUMN PARTITION (key1 = 'val2',
> key2 = 'val2', ...) WITH ('prop_key' = 'prop_val', ...) LOCATION '...';
> ALTER TABLE table_name [IF EXISTS] DROP COLUMN PARTITION (key1
> [=|<|<=|>|>=|!=] 'val1');
> {code}
--
This message was sent by Atlassian JIRA
(v6.2#6252)
