Skip to content

[WIP] HIVE-28931: Convert IN to SEARCH in RexNodeConverter #6495

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
rubenada wants to merge 10 commits into
base: master
Choose a base branch
from
Draft

[WIP] HIVE-28931: Convert IN to SEARCH in RexNodeConverter #6495

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
519f357
HIVE-28931: Convert IN to SEARCH in RexNodeConverter
rubenada May 18, 2026
393009d
Improve RexNodeConverter#rewriteInClause, apply changes on HiveFuncti…
rubenada May 18, 2026
ff031f6
Refactoring
rubenada May 19, 2026
6fc1566
Fix condition
rubenada May 19, 2026
36a0eee
Search can only be created on Literal arguments
rubenada May 20, 2026
a569e23
Fix search handling in HiveReduceExpressionsWithStatsRule, and null h…
rubenada May 20, 2026
3cd2721
Adjust tests
rubenada May 26, 2026
2161927
format
rubenada May 26, 2026
89dbe8a
test: not expanding SEARCH in HiveInBetweenExpandRule
rubenada May 28, 2026
44d0706
rewrite IN into SEARCH in HiveInBetweenExpandRule
rubenada May 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 3 additions & 11 deletions ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/rules/HiveInBetweenExpandRule.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,11 +29,9 @@
import org.apache.calcite.rex.RexCall;
import org.apache.calcite.rex.RexNode;
import org.apache.calcite.rex.RexShuttle;
import org.apache.calcite.rex.RexUnknownAs;
import org.apache.calcite.rex.RexUtil;
import org.apache.calcite.sql.fun.SqlStdOperatorTable;
import org.apache.hadoop.hive.ql.optimizer.calcite.HiveRelFactories;
import org.apache.hadoop.hive.ql.optimizer.calcite.SearchTransformer;
import org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveIn;
import org.apache.hadoop.hive.ql.optimizer.calcite.translator.RexNodeConverter;

Expand Down Expand Up @@ -152,9 +150,6 @@ private RexInBetweenExpander(RexBuilder rexBuilder) {
@Override
public RexNode visitCall(final RexCall call) {
switch (call.getKind()) {
case SEARCH: {
return new SearchTransformer<>(rexBuilder, call, RexUnknownAs.UNKNOWN).transform().accept(this);
}
case AND: {
boolean[] update = {false};
List<RexNode> newOperands = visitList(call.operands, update);
Expand Down Expand Up @@ -182,16 +177,13 @@ public RexNode visitCall(final RexCall call) {
}
default:
if (HiveIn.INSTANCE.equals(call.op)) {
List<RexNode> newOperands = RexNodeConverter.transformInToOrOperands(call.getOperands(), rexBuilder);
if (newOperands == null) {
RexNode newCall = RexNodeConverter.rewriteInClause(call.getOperands(), rexBuilder);
if (newCall == null) {
// We could not execute transformation, return expression
return call;
}
modified = true;
if (newOperands.size() > 1) {
return rexBuilder.makeCall(SqlStdOperatorTable.OR, newOperands);
}
return newOperands.get(0);
return newCall;
}
return super.visitCall(call);
}
Expand Down
17 changes: 15 additions & 2 deletions ...org/apache/hadoop/hive/ql/optimizer/calcite/rules/HiveReduceExpressionsWithStatsRule.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,8 +39,10 @@
import org.apache.calcite.rex.RexUnknownAs;
import org.apache.calcite.rex.RexUtil;
import org.apache.calcite.sql.SqlKind;
import org.apache.calcite.sql.fun.SqlStdOperatorTable;
import org.apache.calcite.util.Pair;
import org.apache.hadoop.hive.ql.optimizer.calcite.RelOptHiveTable;
import org.apache.hadoop.hive.ql.optimizer.calcite.SearchTransformer;
import org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveIn;
import org.apache.hadoop.hive.ql.plan.ColStatistics;
import org.apache.hadoop.hive.ql.stats.StatsUtils;
Expand Down Expand Up @@ -189,10 +191,14 @@ public RexNode visitCall(RexCall call) {
newOperands.add(operand);
}
}
if (newOperands.size() == 1) {
switch (newOperands.size()) {
case 1:
return rexBuilder.makeLiteral(false);
case 2:
return rexBuilder.makeCall(SqlStdOperatorTable.EQUALS, newOperands);
default:
return rexBuilder.makeCall(HiveIn.INSTANCE, newOperands);
}
return rexBuilder.makeCall(HiveIn.INSTANCE, newOperands);
}
} else if (call.getOperands().get(0).getKind() == SqlKind.ROW) {
// Struct
Expand Down Expand Up @@ -269,6 +275,13 @@ public RexNode visitCall(RexCall call) {
}
}
}
} else if (call.getKind() == SqlKind.SEARCH) {
RexNode expanded = new SearchTransformer<>(rexBuilder, call, RexUnknownAs.UNKNOWN).transform();
RexNode processed = expanded.accept(this);
if (expanded != processed) {
return processed;
}
return call;
}

// If we did not reduce, check the children nodes
Expand Down
77 changes: 54 additions & 23 deletions ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/translator/RexNodeConverter.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,10 @@
import org.apache.hadoop.hive.common.type.HiveVarchar;
import org.apache.hadoop.hive.common.type.Timestamp;
import org.apache.hadoop.hive.common.type.TimestampTZ;
import org.apache.hadoop.hive.conf.HiveConf;
import org.apache.hadoop.hive.ql.exec.FunctionRegistry;
import org.apache.hadoop.hive.ql.metadata.Hive;
import org.apache.hadoop.hive.ql.metadata.HiveException;
import org.apache.hadoop.hive.ql.optimizer.calcite.CalciteSemanticException;
import org.apache.hadoop.hive.ql.optimizer.calcite.CalciteSemanticException.UnsupportedFeature;
import org.apache.hadoop.hive.ql.optimizer.calcite.HiveCalciteUtil;
Expand Down Expand Up @@ -117,6 +120,16 @@ public class RexNodeConverter {
private final RexBuilder rexBuilder;
private final RelDataTypeFactory typeFactory;

private static final int MAX_NODES_FOR_IN_TO_OR_TRANSFORMATION;

static {
try {
MAX_NODES_FOR_IN_TO_OR_TRANSFORMATION = HiveConf.getIntVar(
Hive.get().getConf(), HiveConf.ConfVars.HIVEOPT_TRANSFORM_IN_MAXNODES);
} catch (HiveException e) {
throw new IllegalStateException(e);
}
}

/**
* Constructor used by HiveRexExecutorImpl.
Expand Down Expand Up @@ -259,19 +272,12 @@ private RexNode convert(ExprNodeGenericFuncDesc func) throws SemanticException {
// If it is a floor <date> operator, we need to rewrite it
childRexNodeLst = rewriteFloorDateChildren(calciteOp, childRexNodeLst, rexBuilder);
} else if (HiveIn.INSTANCE.equals(calciteOp) && isAllPrimitive) {
if (childRexNodeLst.size() == 2) {
// if it is a single item in an IN clause, transform A IN (B) to A = B
// from IN [A,B] => EQUALS [A,B]
// except complex types
calciteOp = SqlStdOperatorTable.EQUALS;
} else if (RexUtil.isReferenceOrAccess(childRexNodeLst.get(0), true)){
// if it is more than an single item in an IN clause,
// transform from IN [A,B,C] => OR [EQUALS [A,B], EQUALS [A,C]]
// except complex types
// Rewrite to OR is done only if number of operands are less than
// the threshold configured
childRexNodeLst = rewriteInClauseChildren(calciteOp, childRexNodeLst, rexBuilder);
calciteOp = SqlStdOperatorTable.OR;
if (childRexNodeLst.size() == 2 || RexUtil.isReferenceOrAccess(childRexNodeLst.get(0), true)) {
Copy link
Copy Markdown
Member

@zabetak zabetak May 19, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why do we limit the transformation? What are the pros/cons of doing an unconditional rewrite?

Copy link
Copy Markdown
Contributor Author

@rubenada rubenada May 19, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just keeping the original logic to minimize the side effects of the change. I'll investigate why this condition was added in the first place and whether it can be removed...

RexNode rewritten = rewriteInClause(childRexNodeLst, rexBuilder);
assert rewritten instanceof RexCall;
RexCall call = (RexCall) rewritten;
calciteOp = call.op;
childRexNodeLst = call.operands;
}
} else if (calciteOp.getKind() == SqlKind.COALESCE &&
childRexNodeLst.size() > 1) {
Expand Down Expand Up @@ -577,17 +583,42 @@ public static List<RexNode> transformInToOrOperands(List<RexNode> operands, RexB
return disjuncts;
}

public static List<RexNode> rewriteInClauseChildren(SqlOperator op, List<RexNode> childRexNodeLst,
RexBuilder rexBuilder) throws SemanticException {
assert op == HiveIn.INSTANCE;
RexNode firstPred = childRexNodeLst.get(0);
List<RexNode> newChildRexNodeLst = new ArrayList<RexNode>();
for (int i = 1; i < childRexNodeLst.size(); i++) {
newChildRexNodeLst.add(
rexBuilder.makeCall(
SqlStdOperatorTable.EQUALS, firstPred, childRexNodeLst.get(i)));
/**
* This method tries to rewrite IN expression arguments into an equivalent call.
* If there are only two elements, generates an EQUALS:
* IN [A,B] => EQUALS [A,B]
* Otherwise, tries to generate a SEARCH:
* IN [A,B,C] => SEARCH(A, SARG([B..B], [C..C]))
* If this is not possible (e.g., argument types not sufficiently compatible to generate a Calcite SEARCH expression),
* tries to generate an OR expression:
* IN [A,B,C] => OR [EQUALS [A,B], EQUALS [A,C]]
* If this is not possible (e.g., non-deterministic calls are found in the expressions), returns null.
*/
public static RexNode rewriteInClause(List<RexNode> childRexNodeLst, RexBuilder rexBuilder) {
if (childRexNodeLst.size() == 2) {
return rexBuilder.makeCall(SqlStdOperatorTable.EQUALS, childRexNodeLst);
}
Comment on lines +598 to +600
Copy link
Copy Markdown
Member

@zabetak zabetak May 19, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should this become part of RexBuilder#makeIn API if not already the case?

Copy link
Copy Markdown
Contributor Author

@rubenada rubenada May 19, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't know... Calcite's code just creates an IN with one element in this case, which is not incorrect (and I guess it's the expected behavior). I've just kept this special case on Hive's side for the moment to minimize side effects by this refactoring.

return newChildRexNodeLst;

RexNode arg = childRexNodeLst.get(0);
List<RexNode> ranges = childRexNodeLst.subList(1, childRexNodeLst.size());
// Avoid SEARCH on rows for the moment (it can lead to issues in Calcite), and check all types are SEARCH-compatible
if (!arg.getType().isStruct() && ranges.stream().allMatch(range -> range.getKind() == SqlKind.LITERAL
&& !RexLiteral.isNullLiteral(range)
&& SqlTypeUtil.inSameFamily(arg.getType(), range.getType()))) {
RexNode search = rexBuilder.makeIn(arg, ranges);
if (search.getKind() == SqlKind.SEARCH) {
return search;
}
}

// Calcite SEARCH conversion was not possible: generate our own OR expression
if (MAX_NODES_FOR_IN_TO_OR_TRANSFORMATION == 0 || childRexNodeLst.size() <= MAX_NODES_FOR_IN_TO_OR_TRANSFORMATION) {
List<RexNode> newInputs = RexNodeConverter.transformInToOrOperands(childRexNodeLst, rexBuilder);
if (newInputs != null) {
return newInputs.size() == 1 ? newInputs.get(0) : rexBuilder.makeCall(SqlStdOperatorTable.OR, newInputs);
}
}
return null;
}

public static List<RexNode> rewriteCoalesceChildren(
Expand Down
38 changes: 6 additions & 32 deletions ql/src/java/org/apache/hadoop/hive/ql/parse/type/HiveFunctionHelper.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,12 +38,9 @@
import org.apache.calcite.sql.SqlOperator;
import org.apache.calcite.sql.fun.SqlStdOperatorTable;
import org.apache.calcite.util.Util;
import org.apache.hadoop.hive.conf.HiveConf;
import org.apache.hadoop.hive.ql.exec.FunctionInfo;
import org.apache.hadoop.hive.ql.exec.FunctionRegistry;
import org.apache.hadoop.hive.ql.exec.HiveFunctionInfo;
import org.apache.hadoop.hive.ql.metadata.Hive;
import org.apache.hadoop.hive.ql.metadata.HiveException;
import org.apache.hadoop.hive.ql.optimizer.calcite.HiveRexExecutorImpl;
import org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveExtractDate;
import org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveFloorDate;
Expand Down Expand Up @@ -102,16 +99,9 @@ public class HiveFunctionHelper implements FunctionHelper {
private static final Logger LOG = LoggerFactory.getLogger(HiveFunctionHelper.class);

private final RexBuilder rexBuilder;
private final int maxNodesForInToOrTransformation;

public HiveFunctionHelper(RexBuilder rexBuilder) {
this.rexBuilder = rexBuilder;
try {
this.maxNodesForInToOrTransformation = HiveConf.getIntVar(
Hive.get().getConf(), HiveConf.ConfVars.HIVEOPT_TRANSFORM_IN_MAXNODES);
} catch (HiveException e) {
throw new IllegalStateException(e);
}
}

/**
Expand Down Expand Up @@ -267,28 +257,12 @@ public RexNode getExpression(String functionText, FunctionInfo fi,
// If it is a floor <date> operator, we need to rewrite it
inputs = RexNodeConverter.rewriteFloorDateChildren(calciteOp, inputs, rexBuilder);
} else if (HiveIn.INSTANCE.equals(calciteOp)) {
// if it is a single item in an IN clause, transform A IN (B) to A = B
// from IN [A,B] => EQUALS [A,B]
// if it is more than an single item in an IN clause,
// transform from IN [A,B,C] => OR [EQUALS [A,B], EQUALS [A,C]]
// Rewrite to OR is done only if number of operands are less than
// the threshold configured
boolean rewriteToOr = true;
if(maxNodesForInToOrTransformation != 0) {
if(inputs.size() > maxNodesForInToOrTransformation) {
rewriteToOr = false;
}
}
if(rewriteToOr) {
// If there are non-deterministic functions, we cannot perform this rewriting
List<RexNode> newInputs = RexNodeConverter.transformInToOrOperands(inputs, rexBuilder);
if (newInputs != null) {
inputs = newInputs;
if (inputs.size() == 1) {
inputs.add(rexBuilder.makeLiteral(false));
}
calciteOp = SqlStdOperatorTable.OR;
}
RexNode rewritten = RexNodeConverter.rewriteInClause(inputs, rexBuilder);
if (rewritten != null) {
assert rewritten instanceof RexCall;
RexCall call = (RexCall) rewritten;
calciteOp = call.op;
inputs = call.operands;
}
} else if (calciteOp.getKind() == SqlKind.COALESCE &&
inputs.size() > 1) {
Expand Down
12 changes: 6 additions & 6 deletions ql/src/test/results/clientpositive/llap/in_typecheck_char.q.out
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -238,10 +238,10 @@ STAGE PLANS:
Map Operator Tree:
TableScan
alias: ax
filterExpr: (((CAST( s AS STRING) = 'a') and (CAST( t AS STRING) = 'a')) or (null and (CAST( t AS STRING) = 'bb'))) is null (type: boolean)
filterExpr: ((struct(CAST( s AS STRING),CAST( t AS STRING)) = const struct('a','a')) or (struct(s,CAST( t AS STRING)) = const struct(null,'bb'))) is null (type: boolean)
Statistics: Num rows: 3 Data size: 510 Basic stats: COMPLETE Column stats: COMPLETE
Filter Operator
predicate: (((CAST( s AS STRING) = 'a') and (CAST( t AS STRING) = 'a')) or (null and (CAST( t AS STRING) = 'bb'))) is null (type: boolean)
predicate: ((struct(CAST( s AS STRING),CAST( t AS STRING)) = const struct('a','a')) or (struct(s,CAST( t AS STRING)) = const struct(null,'bb'))) is null (type: boolean)
Statistics: Num rows: 1 Data size: 170 Basic stats: COMPLETE Column stats: COMPLETE
Select Operator
Statistics: Num rows: 1 Data size: 170 Basic stats: COMPLETE Column stats: COMPLETE
Expand Down Expand Up @@ -292,7 +292,7 @@ POSTHOOK: query: select 'expected 1',count(*) from ax where ((s,t) in (('a','a')
POSTHOOK: type: QUERY
POSTHOOK: Input: default@ax
#### A masked pattern was here ####
expected 1 1
expected 1 0
PREHOOK: query: explain cbo select count(*) from ax where t in
('a','bb','aa','bbb','ab','ba','aaa','bbb','abc','bc','ac','bca','cab','cb','ca','cbc','z')
PREHOOK: type: QUERY
Expand Down Expand Up @@ -390,7 +390,7 @@ POSTHOOK: Input: default@ax
CBO PLAN:
HiveProject(_c0=[$0])
HiveAggregate(group=[{}], agg#0=[count()])
HiveFilter(condition=[IN($1, _UTF-16LE'a', _UTF-16LE'aa', _UTF-16LE'aaa', _UTF-16LE'ab', _UTF-16LE'abc', _UTF-16LE'ac', _UTF-16LE'ba', _UTF-16LE'bb', _UTF-16LE'bbb', _UTF-16LE'bc', _UTF-16LE'bca', _UTF-16LE'ca', _UTF-16LE'cab', _UTF-16LE'cb', _UTF-16LE'cbc', _UTF-16LE'z')])
HiveFilter(condition=[IN($1, _UTF-16LE'a ', _UTF-16LE'bb ', _UTF-16LE'aa ', _UTF-16LE'bbb ', _UTF-16LE'ab ', _UTF-16LE'ba ', _UTF-16LE'aaa ', _UTF-16LE'bbb ', _UTF-16LE'abc ', _UTF-16LE'bc ', _UTF-16LE'ac ', _UTF-16LE'bca ', _UTF-16LE'cab ', _UTF-16LE'cb ', _UTF-16LE'ca ', _UTF-16LE'cbc ', _UTF-16LE'z ')])
HiveTableScan(table=[[default, ax]], table:alias=[ax])

PREHOOK: query: explain select count(*) from ax where t in
Expand Down Expand Up @@ -419,10 +419,10 @@ STAGE PLANS:
Map Operator Tree:
TableScan
alias: ax
filterExpr: (t) IN ('a ', 'aa ', 'aaa ', 'ab ', 'abc ', 'ac ', 'ba ', 'bb ', 'bbb ', 'bc ', 'bca ', 'ca ', 'cab ', 'cb ', 'cbc ', 'z ') (type: boolean)
filterExpr: (t) IN ('a ', 'bb ', 'aa ', 'bbb ', 'ab ', 'ba ', 'aaa ', 'bbb ', 'abc ', 'bc ', 'ac ', 'bca ', 'cab ', 'cb ', 'ca ', 'cbc ', 'z ') (type: boolean)
Statistics: Num rows: 3 Data size: 255 Basic stats: COMPLETE Column stats: COMPLETE
Filter Operator
predicate: (t) IN ('a ', 'aa ', 'aaa ', 'ab ', 'abc ', 'ac ', 'ba ', 'bb ', 'bbb ', 'bc ', 'bca ', 'ca ', 'cab ', 'cb ', 'cbc ', 'z ') (type: boolean)
predicate: (t) IN ('a ', 'bb ', 'aa ', 'bbb ', 'ab ', 'ba ', 'aaa ', 'bbb ', 'abc ', 'bc ', 'ac ', 'bca ', 'cab ', 'cb ', 'ca ', 'cbc ', 'z ') (type: boolean)
Statistics: Num rows: 3 Data size: 255 Basic stats: COMPLETE Column stats: COMPLETE
Select Operator
Statistics: Num rows: 3 Data size: 255 Basic stats: COMPLETE Column stats: COMPLETE
Expand Down
2 changes: 1 addition & 1 deletion ql/src/test/results/clientpositive/llap/in_typecheck_mixed.q.out
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,4 +80,4 @@ POSTHOOK: query: select 'expected 1',count(*) from ax where ((s,t) in (('a','a')
POSTHOOK: type: QUERY
POSTHOOK: Input: default@ax
#### A masked pattern was here ####
expected 1 1
expected 1 0
16 changes: 8 additions & 8 deletions ql/src/test/results/clientpositive/llap/multi_column_in_single.q.out
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,11 +133,11 @@ STAGE PLANS:
Processor Tree:
TableScan
alias: emps_n7
filterExpr: ((deptno = 2) and ((empno + 1) = 3)) (type: boolean)
filterExpr: (struct((empno + 1),deptno) = const struct(3,2)) (type: boolean)
Filter Operator
predicate: ((deptno = 2) and ((empno + 1) = 3)) (type: boolean)
predicate: (struct((empno + 1),deptno) = const struct(3,2)) (type: boolean)
Select Operator
expressions: empno (type: int), 2 (type: int), empname (type: string)
expressions: empno (type: int), deptno (type: int), empname (type: string)
outputColumnNames: _col0, _col1, _col2
ListSink

Expand All @@ -161,9 +161,9 @@ STAGE PLANS:
Processor Tree:
TableScan
alias: emps_n7
filterExpr: ((deptno <> 2) or ((empno + 1) <> 3)) (type: boolean)
filterExpr: (struct((empno + 1),deptno) <> const struct(3,2)) (type: boolean)
Filter Operator
predicate: ((deptno <> 2) or ((empno + 1) <> 3)) (type: boolean)
predicate: (struct((empno + 1),deptno) <> const struct(3,2)) (type: boolean)
Select Operator
expressions: empno (type: int), deptno (type: int), empname (type: string)
outputColumnNames: _col0, _col1, _col2
Expand All @@ -187,11 +187,11 @@ STAGE PLANS:
Processor Tree:
TableScan
alias: emps_n7
filterExpr: ((deptno = 2) and (((empno * 2) | 1) = (empno + 2))) (type: boolean)
filterExpr: (struct(((empno * 2) | 1),deptno) = struct((empno + 2),2)) (type: boolean)
Filter Operator
predicate: ((deptno = 2) and (((empno * 2) | 1) = (empno + 2))) (type: boolean)
predicate: (struct(((empno * 2) | 1),deptno) = struct((empno + 2),2)) (type: boolean)
Select Operator
expressions: empno (type: int), 2 (type: int), empname (type: string)
expressions: empno (type: int), deptno (type: int), empname (type: string)
outputColumnNames: _col0, _col1, _col2
ListSink

Expand Down
Loading
Loading