[
https://issues.apache.org/jira/browse/IGNITE-24664?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17932171#comment-17932171
]
Aleksey Plekhanov commented on IGNITE-24664:
--------------------------------------------
Local performance test:
Before fix:
{noformat}
Benchmark (engine) Mode Cnt Score Error Units
JmhSqlBenchmark.querySimpleUnique H2 thrpt 3 192,254 ± 3,643 ops/s
JmhSqlBenchmark.querySimpleUnique CALCITE thrpt 3 147,601 ± 28,635 ops/s
{noformat}
After fix:
{noformat}
Benchmark (engine) Mode Cnt Score Error Units
JmhSqlBenchmark.querySimpleUnique H2 thrpt 3 193,708 ± 23,837 ops/s
JmhSqlBenchmark.querySimpleUnique CALCITE thrpt 3 160,935 ± 10,954 ops/s
{noformat}
> Calcite engine. Use _key_PK index scan instead of hash index scan if possible
> -----------------------------------------------------------------------------
>
> Key: IGNITE-24664
> URL: https://issues.apache.org/jira/browse/IGNITE-24664
> Project: Ignite
> Issue Type: Improvement
> Reporter: Aleksey Plekhanov
> Assignee: Aleksey Plekhanov
> Priority: Major
> Labels: calcite, ise
> Time Spent: 20m
> Remaining Estimate: 0h
>
> Looks like _key_PK index scan is more effective than hash index scan. Hash
> index scan requires also entity type check, but _key_PK index scan only built
> for required entity type, so type check can be skipped.
> H2 engine uses PK index scan for table scans (see H2TableScanIndex.delegate()
> method).
> Calcite engine should also use PK index scan when indexes are in correct
> state.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
