[
https://issues.apache.org/jira/browse/PHOENIX-7258?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17824751#comment-17824751
]
ASF GitHub Bot commented on PHOENIX-7258:
-----------------------------------------
sanjeet006py commented on code in PR #1851:
URL: https://github.com/apache/phoenix/pull/1851#discussion_r1517610019
##########
phoenix-core-client/src/main/java/org/apache/phoenix/optimize/QueryOptimizer.java:
##########
@@ -247,6 +248,10 @@ private List<QueryPlan>
getApplicablePlansForSingleFlatQuery(QueryPlan dataPlan,
}
plans.add(0, hintedPlan);
}
+ if (dataPlan.getContext().getScanRanges().isPointLookup()
+ && stopAtBestPlan && dataPlan.isApplicable()) {
+ return Collections.<QueryPlan> singletonList(dataPlan);
+ }
Review Comment:
@tkhurana I see that if we remove this block then code goes on to create
query plan for each index of the original data table. And, to create query plan
for each index at least 1 RPC call is done so, I think this can be expensive
when the end result we know is original data plan but we will still do RPC
calls for other indexes.
> Query Optimizer should pick Index hint even for point lookup queries
> --------------------------------------------------------------------
>
> Key: PHOENIX-7258
> URL: https://issues.apache.org/jira/browse/PHOENIX-7258
> Project: Phoenix
> Issue Type: Improvement
> Affects Versions: 5.2.0, 5.1.3
> Reporter: Viraj Jasani
> Assignee: Sanjeet Malhotra
> Priority: Major
>
> For better performance, user can create covered indexes such that the indexed
> columns are same as composite primary key of the data table, but with
> different order. For instance, create data table with columns PK1, PK2, PK3,
> C1, C2 columns with primary key being PK1, Pk2, PK3. In order to get better
> performance from HBase block caching, if the data with same PK3 values are
> going to reside as close to each other as possible, we can create an index on
> PK3, PK2, PK1 and also include columns C1 and C2.
> For point lookups on the data table, it might still be helpful to query index
> table depending on the usecase. We should allow using index hint to query the
> index table for point lookup.
> When the query optimizer identifies that the query is point lookup for the
> data table and if the "stop at best plan" is true, then it immediately
> returns without checking the hint. We should check for hint and if applicable
> index based hint plan is identified, use it.
>
> Assuming getHintedPlanIfApplicable() retrieves hinted plan, it can look
> something like:
> {code:java}
> if (dataPlan.getContext().getScanRanges().isPointLookup() && stopAtBestPlan
> && dataPlan.isApplicable()) {
> if (indexes.isEmpty() || select.getHint().getHint(Hint.INDEX) == null) {
> return Collections.singletonList(dataPlan);
> }
> QueryPlan hintedPlan = getHintedPlanIfApplicable(dataPlan, statement,
> targetColumns,
> parallelIteratorFactory, select, indexes);
> if (hintedPlan != null) {
> PTable index = hintedPlan.getTableRef().getTable();
> if (hintedPlan.isApplicable() && (index.getIndexWhere() == null
> || isPartialIndexUsable(select, dataPlan, index))) {
> return Collections.singletonList(hintedPlan);
> }
> }
> return Collections.singletonList(dataPlan);
> } {code}
> We still need to be optimal i.e. if the hinted index plan is not applicable
> or useful, we still need to immediately return the data plan of point lookup.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
