[
https://issues.apache.org/jira/browse/DRILL-7450?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16981365#comment-16981365
]
ASF GitHub Bot commented on DRILL-7450:
---------------------------------------
vvysotskyi commented on pull request #1907: DRILL-7450: Improve performance for
ANALYZE command
URL: https://github.com/apache/drill/pull/1907
Jira - [DRILL-7450](https://issues.apache.org/jira/browse/DRILL-7450).
This pull request contains several commits that may be reviewed one-by-one.
The first commit "Improve performance for `ANALYZE` command" contains
changes for two-phase metadata stream aggregation support. The main logic is
placed in `MetadataAggPrule` - it converts the lowest `MetadataAggRel` into the
following plan:
```
MetadataAggPrel(PHASE_2of2)
\
HashToMergeExchangePrel
\
MetadataAggPrel(PHASE_1of2)
```
Also, this commit introduces `ConvertMetadataAggregateToDirectScanRule` rule
which replaces the lowest `MetadataAggRel` with `DrillDirectScanRel` populated
from row groups metadata if required metadata is present:
```
MetadataAggRel(metadataLevel=ROW_GROUP)
\
DrillScanRel
```
is converted into
```
DrillDirectScanRel
```
The last commit contains changes to use hash aggregation for `PHASE_1of2`
for `ANALYZE` to reduce memory usage and avoid sorting non-aggregated data.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> Improve performance for ANALYZE command
> ---------------------------------------
>
> Key: DRILL-7450
> URL: https://issues.apache.org/jira/browse/DRILL-7450
> Project: Apache Drill
> Issue Type: Bug
> Affects Versions: 1.17.0
> Reporter: Vova Vysotskyi
> Assignee: Vova Vysotskyi
> Priority: Blocker
> Fix For: 1.17.0
>
>
> In the scope of DRILL-7273 was introduced ANALYZE command for collecting
> metadata and storing it to the metastore.
> But current implementation uses too much memory and is low-performant. It
> uses stream aggregate for collecting metadata, but all incoming data should
> be sorted before producing the aggregation. Memory usage may be reduced by
> using hash aggregate, so sort may not be produced and it should increase
> performance.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
