[CALCITE-4281] fix getMaxRowCount for RelSubset to return reasonable values

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.25.0
• Component/s: core
• Assignee: Unassigned
• Reporter: chunfeng pei
• Created: 25/Sep/20 23:59
• Updated: 14/Oct/24 12:44

Description

currently getMaxRowCount(RelSubset rel, RelMetadataQuery mq) only have a short term fix for CALCITE-1018.  https://issues.apache.org/jira/browse/CALCITE-1018

we need a complete solution for this, as suggested in https://issues.apache.org/jira/browse/CALCITE-1048

quotes: Clearly RelMdMaxRowCount of a RelSubset should be the lowest max-row-count of any relational expression in that subset. Each kind of metadata should have a way of deriving a subset's value from the constituent rel values; in the case of RelMdMaxRowCount the roll-up function is "min".

also a sample solution provided by weijie : 
public Double getMaxRowCount(RelSubset rel, RelMetadataQuery mq) { Double lowest = Double.POSITIVE_INFINITY; for (RelNode node : rel.getRels()) { Double maxRowCount = mq.getMaxRowCount(node); if (maxRowCount == null)

{ continue; }

if (lowest.compareTo(maxRowCount) > 0) {
lowest = maxRowCount;
}
} return lowest;
}
then comments from jhyde :

quote: I was thinking of doing the same thing in a more abstract way, namely providing a "fold" function.

this ticket is opened to address this issue. this is also related to the following ticket:

https://issues.apache.org/jira/browse/CALCITE-3963

Issue Links

• Blocked: CALCITE-6586 Some Rules not firing due to RelMdPredicates returning null in VolcanoPlanner
• is related to: CALCITE-1048 Make metadata more robust
• is related to: CALCITE-1018 SortJoinTransposeRule not firing due to getMaxRowCount(RelSubset) returning null