[
https://issues.apache.org/jira/browse/PIG-4107?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14087938#comment-14087938
]
Julian Hyde commented on PIG-4107:
----------------------------------
A PoC would look like Pig, and use Pig's parser and metadata (HCat), run Pig's
optimization rules, then translate Pig's AST to Optiq RelNodes. Then Optiq
logical and physical rules could kick in, including stuff like materialized
view matching.
Code would be part of Pig, dependent on Optiq. (There is code we would like to
use at some point, but not for the PoC, that is in the Hive code base.)
In the PoC, we would generate Optiq enumerable plans. But potentially the
system could use any back-end supported by Optiq, including Tez (via the
HiveXxxRel physical algebra).
Challenges include: Pig's type system is different from the Optiq/Hive/SQL type
system; Pig has a split operator, and operators can have multiple outputs;
Pig's UDFs are different. But these challenges don't need to be addressed in
the initial version.
> Optiq for query optimization
> ----------------------------
>
> Key: PIG-4107
> URL: https://issues.apache.org/jira/browse/PIG-4107
> Project: Pig
> Issue Type: New Feature
> Components: impl
> Reporter: Daniel Dai
>
> Optiq is a query planning engine which is currently in Apache incubator. We'd
> like to explore the possibility to use Optiq to do query optimization,
> material view generation for Pig.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
