[
https://issues.apache.org/jira/browse/HIVE-22893?focusedWorklogId=393331&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-393331
]
ASF GitHub Bot logged work on HIVE-22893:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 26/Feb/20 10:11
Start Date: 26/Feb/20 10:11
Worklog Time Spent: 10m
Work Description: kgyrtkirk commented on pull request #915: HIVE-22893
StatEstimate
URL: https://github.com/apache/hive/pull/915#discussion_r384342676
##########
File path: ql/src/java/org/apache/hadoop/hive/ql/stats/StatsUtils.java
##########
@@ -1560,6 +1554,32 @@ public static ColStatistics
getColStatisticsFromExpression(HiveConf conf, Statis
}
}
+ if (conf.getBoolVar(ConfVars.HIVE_STATS_USE_UDF_ESTIMATORS)) {
+ Optional<IStatEstimatorProvider> sep =
engfd.getGenericUDF().adapt(IStatEstimatorProvider.class);
+ if (sep.isPresent()) {
+ Optional<IStatEstimator> se = sep.get().getStatEstimator();
+ if (se.isPresent()) {
+ List<ColStatistics> csList = new ArrayList<ColStatistics>();
+ for (ExprNodeDesc child : engfd.getChildren()) {
+ ColStatistics cs = getColStatisticsFromExpression(conf,
parentStats, child);
+ if (cs == null) {
+ break;
+ }
+ csList.add(cs);
+ }
+ if(csList.size() == engfd.getChildren().size()) {
+ Optional<ColStatistics> res = se.get().estimate(csList);
+ if (res.isPresent()) {
Review comment:
> When could this happen?
For example if an estimator targets only cases in which a specific argument
is a constant.
Right now I don't think there are any specific cases in which the estimation
might not be available; however I think it's ok for an estimator to decide to
not cover some cases...
instead of a probably missing nullcheck later; the optional is already
checked now...
----------------------------------------------------------------
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]
Issue Time Tracking
-------------------
Worklog Id: (was: 393331)
> Enhance data size estimation for fields computed by UDFs
> --------------------------------------------------------
>
> Key: HIVE-22893
> URL: https://issues.apache.org/jira/browse/HIVE-22893
> Project: Hive
> Issue Type: Improvement
> Components: Statistics
> Reporter: Zoltan Haindrich
> Assignee: Zoltan Haindrich
> Priority: Major
> Labels: pull-request-available
> Attachments: HIVE-22893.01.patch, HIVE-22893.02.patch,
> HIVE-22893.03.patch, HIVE-22893.04.patch, HIVE-22893.05.patch,
> HIVE-22893.06.patch, HIVE-22893.07.patch, HIVE-22893.08.patch,
> HIVE-22893.09.patch, HIVE-22893.10.patch, HIVE-22893.11.patch,
> HIVE-22893.12.patch, HIVE-22893.13.patch
>
> Time Spent: 1h 20m
> Remaining Estimate: 0h
>
> Right now if we have columnstat on a column ; we use that to estimate things
> about the column; - however if an UDF is executed on a column ; the resulting
> column is treated as unknown thing and defaults are assumed.
> An improvement could be to give wide estimation(s) in case of frequently used
> udf.
> For example; consider {{substr(c,1,1)}} ; no matter what the input; the
> output is at most a 1 long string
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
