[
https://issues.apache.org/jira/browse/CALCITE-3122?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16885464#comment-16885464
]
Khai Tran commented on CALCITE-3122:
------------------------------------
Hi [~zabetak], thanks a lot for your suggestions.
In this feature, I get the Pig logical plan from Pig parser then traverse
through this plan, use RelBuilder to construct Calcite logical plan, let's call
it plan #1. After this step, I need to write an additional rule to optimize
plan #1 into plan #2. As an example, Pig group-by + aggregate is literally
translated as a Calcite aggregate with COLLECT() agg func, then applying a
Project that use Pig aggregate UDFs (work on Pig DataBag or SQL multiset, which
is result of COLLECT()). So the rule will convert COLLECT() + Pig aggregate
UDFs into Calcite builtin aggregate operator.
And we have other use cases that need to optimize a RelNode plan using a given
set of rules and probably using a customer planner to set costs in the way that
can enforce certain rules. So coupling RelNode with planners make it hard to do
so.
I will try to check out more about HepPlanner, but do you have any example of
setting HepPlanner for RelBuilder?
> Convert Pig Latin scripts into Calcite logical plan
> ----------------------------------------------------
>
> Key: CALCITE-3122
> URL: https://issues.apache.org/jira/browse/CALCITE-3122
> Project: Calcite
> Issue Type: New Feature
> Components: core, piglet
> Reporter: Khai Tran
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.21.0
>
> Time Spent: 1h 10m
> Remaining Estimate: 0h
>
> We create an internal Calcite repo at LinkedIn and develop APIs to parse any
> Pig Latin scripts into Calcite logical plan. The code was tested in nearly
> ~1000 Pig scripts written at LinkedIn.
> Changes:
> 1. piglet: main conversion code live there, include:
> * APIs to convert any Pig scripts into RelNode plans or SQL statements
> * Use Pig Grunt parser to parse Pig Latin scripts into Pig logical plan
> (DAGs)
> * Convert Pig schemas into RelDatatype
> * Traverse through Pig expression plan and convert Pig expressions into
> RexNodes
> * Map some basic Pig UDFs to Calcite SQL operators
> * Build Calcite UDFs for any other Pig UDFs, including UDFs written in both
> Java and Python
> * Traverse (DFS) through Pig logical plans to convert each Pig logical nodes
> to RelNodes
> * Have an optimizer rule to optimize Pig group/cogroup into Aggregate
> operators
> 2. core:
> * Implement other RelNode in Rel2Sql so that Pig can be translated into SQL
> * Other minor changes in a few other classes to make Pig to Calcite works
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
