[
https://issues.apache.org/jira/browse/CALCITE-7053?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17986812#comment-17986812
]
Wenzhuang Zhu commented on CALCITE-7053:
----------------------------------------
Actually, depth-first requires garbage collection if the {{getParent}} call
affects the {{HepPlanner}} results.
I fixed this problem in our internal system, but I haven't have time to share
the code with the community over the past three months.
{*}{*}I will share the complete solution later at
https://issues.apache.org/jira/browse/CALCITE-6907 if there are more people
working on HepPlanner latency for large plan.
> In HepPlanner, move down the collectGarbage to topological order for better
> optimizer performance
> -------------------------------------------------------------------------------------------------
>
> Key: CALCITE-7053
> URL: https://issues.apache.org/jira/browse/CALCITE-7053
> Project: Calcite
> Issue Type: Improvement
> Reporter: xiong duan
> Assignee: xiong duan
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.41.0
>
>
> According to the java comment in HepPlanner#getGraphIterator:
> {code:java}
> // Make sure there's no garbage, because topological sort
> // doesn't start from a specific root, and rules can't
> // deal with firing on garbage.
> // FIXME jvs 25-Sept-2006: I had to move this earlier because
> // of FRG-215, which is still under investigation. Once we
> // figure that one out, move down to location below for
> // better optimizer performance.
> {code}
> Based on the current implementation, I haven't found the reason why the
> depth-first order requires garbage collection every time. Since it always
> starts the search from a specific
> Vertex, the issue mentioned the FRG-215 -
> I haven't found any description or information about this problem.
> Therefore, this PR attempts to delegate the collectGarbage operation to the
> topological order.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
