[
https://issues.apache.org/jira/browse/HIVE-28252?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated HIVE-28252:
----------------------------------
Labels: pull-request-available (was: )
> AssertionError when using HiveTableScan with a HepPlanner cluster
> -----------------------------------------------------------------
>
> Key: HIVE-28252
> URL: https://issues.apache.org/jira/browse/HIVE-28252
> Project: Hive
> Issue Type: Improvement
> Components: CBO, Tests
> Affects Versions: 4.0.0
> Reporter: Stamatis Zampetakis
> Assignee: Stamatis Zampetakis
> Priority: Major
> Labels: pull-request-available
>
> The {{HiveTableScan}} operator throws an
> [AssertionError|https://github.com/apache/hive/blob/7950967eae9640fcc0aa22f4b6c7906b34281eac/ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/reloperators/HiveTableScan.java#L153]
> if the operator does not have the {{HiveRelNode.CONVENTION}} set.
> The {{HepPlanner}} does not use any
> [RelTraitDef|https://github.com/apache/calcite/blob/f854ef5ee480e0ff893b18d27ec67dc381ee2244/core/src/main/java/org/apache/calcite/plan/AbstractRelOptPlanner.java#L276]
> so the default [empty traitset for the respective
> cluster|https://github.com/apache/calcite/blob/f854ef5ee480e0ff893b18d27ec67dc381ee2244/core/src/main/java/org/apache/calcite/plan/RelOptCluster.java#L99]
> is gonna be always empty.
> In principle we should not be able to use the {{HiveTableScan}} operator with
> {{HepPlanner}}. However, the optimizer heavily uses the {{HepPlanner}} (in
> fact more than the {{VolcanoPlanner}} and it is reasonable to wonder how is
> this possible given that this assertion is in place. The assertion is
> circumvented by creating a cluster from a
> [VolcanoPlanner|https://github.com/apache/hive/blob/7950967eae9640fcc0aa22f4b6c7906b34281eac/ql/src/java/org/apache/hadoop/hive/ql/parse/CalcitePlanner.java#L1620]
> and then using it in the
> [HepPlanner|https://github.com/apache/hive/blob/7950967eae9640fcc0aa22f4b6c7906b34281eac/ql/src/java/org/apache/hadoop/hive/ql/parse/CalcitePlanner.java#L2422].
>
> This cluster usage is a bit contrived but does not necessarily need to change
> at this stage.
> Nevertheless, since the {{HiveTableScan}} operator is suitable to run with
> the {{HepPlanner}} the assertion can be relaxed (or removed altogether) to
> better reflect the actual usage of the operator, and allow passing a "true"
> HepPlanner cluster inside the operator.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
