[
https://issues.apache.org/jira/browse/IMPALA-15097?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18088541#comment-18088541
]
Aleksandr Efimov commented on IMPALA-15097:
-------------------------------------------
Posted a patch:
http://gerrit.cloudera.org:8080/24451
The patch marks runtime-filter consumer edges as dashed when scan-node
runtime filter counters show processed input but zero rejected input. At
higher verbosity it also shows compact rejected/total counters on the graph.
One implementation detail: per-instance scan-node counters are summed when
available, while Averaged Fragment counters are used only as fallback. This
keeps the graph values as total skipped input and avoids double-counting.
Validation:
- gerrit-auto-critic dryrun passes locally
- focused plan-graph unit tests pass
- generated SVGs from the attached profile with Graphviz
> Show runtime filter effectiveness in plan-graph.py
> --------------------------------------------------
>
> Key: IMPALA-15097
> URL: https://issues.apache.org/jira/browse/IMPALA-15097
> Project: IMPALA
> Issue Type: Improvement
> Reporter: Quanlong Huang
> Assignee: Aleksandr Efimov
> Priority: Major
> Attachments: query-plan1.svg, rf_profile.txt
>
>
> When verbosity is larger than 0, plan-graph.py generates nodes and lines for
> runtime filters. It'd be helpful to show ineffective runtime filters, e.g.,
> by using dash lines.
> It'd be more helpful if we should how many files/rows, etc. are skipped by
> the filter.
> Uploaded an example profile and the generated svg file. The command is
> {code:java}
> bin/diagnostics/experimental/plan-graph.py -v1 rf_profile.txt | dot -T svg -o
> query-plan1.svg {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
