Support LIKE expressions in filtering queries

patch by Pranav Shenoy; reviewed by Caleb Rackliffe and David Capwell for CASSANDRA-17198
This commit is contained in:
Pranav Shenoy 2025-09-08 20:43:07 -07:00 committed by Caleb Rackliffe
parent 06440e947a
commit cf806cac1a
12 changed files with 156 additions and 41 deletions

View File

@ -1,4 +1,5 @@
5.1
* Support LIKE expressions in filtering queries (CASSANDRA-17198)
* Make legacy index rebuilds safe on Gossip -> TCM upgrades (CASSANDRA-20887)
* Minor improvements and hardening for IndexHints (CASSANDRA-20888)
* Stop repair scheduler if two major versions are detected (CASSANDRA-20048)

View File

@ -611,7 +611,7 @@ public enum Operator
public boolean isSatisfiedBy(AbstractType<?> type, ByteBuffer leftOperand, ByteBuffer rightOperand)
{
return leftOperand != null && ByteBufferUtil.contains(leftOperand, rightOperand);
return leftOperand != null && leftOperand.equals(rightOperand);
}
},
LIKE(14)
@ -621,12 +621,6 @@ public enum Operator
{
throw new UnsupportedOperationException();
}
@Override
public boolean requiresIndexing()
{
return true;
}
},
ANN(15)
{

View File

@ -105,7 +105,8 @@ public class Ordering
{
return new SimpleRestriction(ColumnsExpression.singleColumn(columnMetadata, tableMetadata),
Operator.ANN,
Terms.of(vectorValue));
Terms.of(vectorValue),
false);
}
}

View File

@ -187,7 +187,7 @@ public final class Relation
* @return the <code>Restriction</code> corresponding to this <code>Relation</code>
* @throws InvalidRequestException if this <code>Relation</code> is not valid
*/
public SingleRestriction toRestriction(TableMetadata table, VariableSpecifications boundNames)
public SingleRestriction toRestriction(TableMetadata table, VariableSpecifications boundNames, boolean allowFiltering)
{
ColumnsExpression columnsExpression = rawExpressions.prepare(table);
@ -217,9 +217,9 @@ public final class Relation
// An IN restriction with only one element is the same as an EQ restriction
if (operator.isIN() && terms.containsSingleTerm())
return new SimpleRestriction(columnsExpression, Operator.EQ, terms);
return new SimpleRestriction(columnsExpression, Operator.EQ, terms, allowFiltering);
return new SimpleRestriction(columnsExpression, operator, terms);
return new SimpleRestriction(columnsExpression, operator, terms, allowFiltering);
}
public ColumnIdentifier column()

View File

@ -43,6 +43,7 @@ import org.apache.cassandra.utils.ByteBufferUtil;
import static org.apache.cassandra.cql3.statements.RequestValidations.checkFalse;
import static org.apache.cassandra.cql3.statements.RequestValidations.invalidRequest;
import java.util.Optional;
/**
* A simple predicate on a columns expression (e.g. columnA = X).
@ -64,11 +65,17 @@ public final class SimpleRestriction implements SingleRestriction
*/
private final Terms values;
public SimpleRestriction(ColumnsExpression columnsExpression, Operator operator, Terms values)
/**
* Indicates if the query has allow filtering
*/
private final boolean allowFiltering;
public SimpleRestriction(ColumnsExpression columnsExpression, Operator operator, Terms values, boolean allowFiltering)
{
this.columnsExpression = columnsExpression;
this.operator = operator;
this.values = values;
this.allowFiltering = allowFiltering;
}
@Override
@ -344,11 +351,11 @@ public final class SimpleRestriction implements SingleRestriction
else if (operator == Operator.LIKE)
{
LikePattern pattern = LikePattern.parse(buffers.get(0));
// there must be a suitable INDEX for LIKE_XXX expressions
RowFilter.SimpleExpression expression = filter.add(column, pattern.kind().operator(), pattern.value());
indexRegistry.getBestIndexFor(expression, indexHints)
.orElseThrow(() -> invalidRequest("%s is only supported on properly indexed columns",
expression));
Optional<Index> index = indexRegistry.getBestIndexFor(expression, indexHints);
if(!index.isPresent() && !allowFiltering)
throw invalidRequest("%s is only supported on properly indexed columns or with ALLOW FILTERING", expression);
}
else
{

View File

@ -226,11 +226,11 @@ public final class StatementRestrictions
if (!forView)
throw new InvalidRequestException("Unsupported restriction: " + relation);
this.notNullColumns.addAll(relation.toRestriction(table, boundNames).columns());
this.notNullColumns.addAll(relation.toRestriction(table, boundNames, allowFiltering).columns());
}
else if (operator.requiresIndexing())
{
Restriction restriction = relation.toRestriction(table, boundNames);
Restriction restriction = relation.toRestriction(table, boundNames, allowFiltering);
if (!type.allowUseOfSecondaryIndices() || !restriction.hasSupportingIndex(indexRegistry, indexHints))
throw invalidRequest("%s restriction is only supported on properly " +
@ -240,7 +240,7 @@ public final class StatementRestrictions
}
else
{
addRestriction(relation.toRestriction(table, boundNames), indexRegistry, indexHints);
addRestriction(relation.toRestriction(table, boundNames, allowFiltering), indexRegistry, indexHints);
}
}

View File

@ -31,6 +31,7 @@ import org.apache.cassandra.db.marshal.ListType;
import org.apache.cassandra.index.sai.StorageAttachedIndex;
import org.apache.cassandra.index.sai.analyzer.AbstractAnalyzer;
import org.apache.cassandra.index.sai.utils.IndexTermType;
import org.apache.cassandra.utils.ByteBufferUtil;
/**
* An {@link Expression} is an internal representation of an index query operation. They are built from
@ -79,7 +80,7 @@ public abstract class Expression
public enum IndexOperator
{
EQ, RANGE, CONTAINS_KEY, CONTAINS_VALUE, ANN, IN;
EQ, RANGE, CONTAINS_KEY, CONTAINS_VALUE, ANN, IN, LIKE_PREFIX, LIKE_SUFFIX, LIKE_MATCHES, LIKE_CONTAINS;
public static IndexOperator valueOf(Operator operator)
{
@ -106,6 +107,15 @@ public abstract class Expression
case IN:
return IN;
case LIKE_PREFIX:
return LIKE_PREFIX;
case LIKE_SUFFIX:
return LIKE_SUFFIX;
case LIKE_CONTAINS:
return LIKE_CONTAINS;
case LIKE_MATCHES:
return LIKE_MATCHES;
default:
return null;
@ -114,13 +124,18 @@ public abstract class Expression
public boolean isEquality()
{
return this == EQ || this == CONTAINS_KEY || this == CONTAINS_VALUE || this == IN;
return this == EQ || this == CONTAINS_KEY || this == CONTAINS_VALUE || this == IN || isLikeVariant();
}
public boolean isEqualityOrRange()
{
return isEquality() || this == RANGE;
}
public boolean isLikeVariant()
{
return this == LIKE_SUFFIX || this == LIKE_PREFIX || this == LIKE_CONTAINS || this == LIKE_MATCHES;
}
}
public abstract boolean isNotIndexed();
@ -172,6 +187,10 @@ public abstract class Expression
case EQ:
case CONTAINS:
case CONTAINS_KEY:
case LIKE_PREFIX:
case LIKE_SUFFIX:
case LIKE_MATCHES:
case LIKE_CONTAINS:
case IN:
lower = new Bound(value, indexTermType, true);
upper = lower;
@ -354,6 +373,26 @@ public abstract class Expression
}
}
break;
case LIKE_PREFIX:
{
isMatch = ByteBufferUtil.startsWith(term, requestedValue);
break;
}
case LIKE_SUFFIX:
{
isMatch = ByteBufferUtil.endsWith(term, requestedValue);
break;
}
case LIKE_CONTAINS:
{
isMatch = ByteBufferUtil.contains(term, requestedValue);
break;
}
case LIKE_MATCHES:
{
isMatch = term.equals(requestedValue);
break;
}
}
return isMatch;
}

View File

@ -1645,7 +1645,7 @@ public class ClusteringColumnRestrictionsTest
private static Restriction newSingleRestriction(TableMetadata tableMetadata, int index, Operator operator, ByteBuffer... values)
{
ColumnMetadata column = getClusteringColumnDefinition(tableMetadata, index);
return new SimpleRestriction(ColumnsExpression.singleColumn(column, tableMetadata), operator, toTerms(values));
return new SimpleRestriction(ColumnsExpression.singleColumn(column, tableMetadata), operator, toTerms(values), false);
}
/**
@ -1669,7 +1669,8 @@ public class ClusteringColumnRestrictionsTest
TupleType tupleType = new TupleType(types);
return new SimpleRestriction(ColumnsExpression.multiColumns(columns, tableMetadata),
Operator.EQ,
Terms.of(new MultiElements.Value(tupleType, asList(values))));
Terms.of(new MultiElements.Value(tupleType, asList(values))),
false);
}
/**
@ -1700,7 +1701,7 @@ public class ClusteringColumnRestrictionsTest
{
terms.add(new MultiElements.Value(tupleType, values[i]));
}
return new SimpleRestriction(ColumnsExpression.multiColumns(columns, tableMetadata), Operator.IN, Terms.of(terms));
return new SimpleRestriction(ColumnsExpression.multiColumns(columns, tableMetadata), Operator.IN, Terms.of(terms), false);
}
/**
@ -1737,7 +1738,8 @@ public class ClusteringColumnRestrictionsTest
TupleType type = new TupleType(types);
return new SimpleRestriction(ColumnsExpression.multiColumns(columns, tableMetadata),
operator,
Terms.of(new MultiElements.Value(type, asList(values))));
Terms.of(new MultiElements.Value(type, asList(values))),
false);
}
/**

View File

@ -843,16 +843,16 @@ public class SecondaryIndexTest extends CQLTester
// LIKE is not supported on indexes of non-literal values
// this is rejected before binding, so the value isn't available in the error message
assertInvalidMessage("LIKE restriction is only supported on properly indexed columns. v3 LIKE ? is not valid",
assertInvalidMessage(StatementRestrictions.REQUIRES_ALLOW_FILTERING_MESSAGE,
"SELECT * FROM %s WHERE v3 LIKE ?",
"%abc");
assertInvalidMessage("LIKE restriction is only supported on properly indexed columns. v3 LIKE ? is not valid",
assertInvalidMessage(StatementRestrictions.REQUIRES_ALLOW_FILTERING_MESSAGE,
"SELECT * FROM %s WHERE v3 LIKE ?",
"%abc%");
assertInvalidMessage("LIKE restriction is only supported on properly indexed columns. v3 LIKE ? is not valid",
assertInvalidMessage(StatementRestrictions.REQUIRES_ALLOW_FILTERING_MESSAGE,
"SELECT * FROM %s WHERE v3 LIKE ?",
"%abc%");
assertInvalidMessage("LIKE restriction is only supported on properly indexed columns. v3 LIKE ? is not valid",
assertInvalidMessage(StatementRestrictions.REQUIRES_ALLOW_FILTERING_MESSAGE,
"SELECT * FROM %s WHERE v3 LIKE ?",
"abc");
}

View File

@ -33,7 +33,7 @@ import org.apache.cassandra.service.StorageService;
import static java.lang.String.format;
import static org.assertj.core.api.Assertions.assertThatThrownBy;
import static org.junit.Assert.assertNotNull;
import org.apache.cassandra.exceptions.InvalidRequestException;
/**
* Tests that {@code ALLOW FILTERING} is required only if needed.
*/
@ -433,4 +433,83 @@ public class AllowFilteringTest extends SAITester
execute("SELECT * FROM %s WHERE v=0");
execute("SELECT * FROM %s WHERE v=0 ALLOW FILTERING");
}
@Test
public void testAllowFilteringWithLikePrefixPostFiltering()
{
createTable("CREATE TABLE %S (k1 int, k2 text, k3 int, PRIMARY KEY (k1))");
createIndex("CREATE INDEX ON %s(k3) USING 'sai'");
execute("insert into %s (k1, k2, k3) values (1, 'fo', 1)");
execute("insert into %s (k1, k2, k3) values (2, 'foo', 2)");
execute("insert into %s (k1, k2, k3) values (3, 'fo', 3)");
execute("insert into %s (k1, k2, k3) values (4, 'ba', 4)");
execute("insert into %s (k1, k2, k3) values (5, 'bar', 5)");
assertRowCount(execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE 'f%%' ALLOW FILTERING"), 3);
assertRowCount(execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE 'ca%%' ALLOW FILTERING"), 0);
assertThatThrownBy(() -> execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE 'f%%'"))
.hasMessage(StatementRestrictions.REQUIRES_ALLOW_FILTERING_MESSAGE)
.isInstanceOf(InvalidRequestException.class);
}
@Test
public void testAllowFilteringWithLikeSuffixPostFiltering()
{
createTable("CREATE TABLE %S (k1 int, k2 text, k3 int, PRIMARY KEY (k1))");
createIndex("CREATE INDEX ON %s(k3) USING 'sai'");
execute("insert into %s (k1, k2, k3) values (1, 'fo', 1)");
execute("insert into %s (k1, k2, k3) values (2, 'foo', 2)");
execute("insert into %s (k1, k2, k3) values (3, 'fo', 3)");
execute("insert into %s (k1, k2, k3) values (4, 'ba', 4)");
execute("insert into %s (k1, k2, k3) values (5, 'bar', 5)");
assertRowCount(execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE '%%o' ALLOW FILTERING"), 3);
assertRowCount(execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE '%%c' ALLOW FILTERING"), 0);
assertThatThrownBy(() -> execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE '%%c'"))
.hasMessage(StatementRestrictions.REQUIRES_ALLOW_FILTERING_MESSAGE)
.isInstanceOf(InvalidRequestException.class);
}
@Test
public void testAllowFilteringWithLikeContainsPostFiltering()
{
createTable("CREATE TABLE %S (k1 int, k2 text, k3 int, PRIMARY KEY (k1))");
createIndex("CREATE INDEX ON %s(k3) USING 'sai'");
execute("insert into %s (k1, k2, k3) values (1, 'fo', 1)");
execute("insert into %s (k1, k2, k3) values (2, 'foo', 2)");
execute("insert into %s (k1, k2, k3) values (3, 'fo', 3)");
execute("insert into %s (k1, k2, k3) values (4, 'ba', 4)");
execute("insert into %s (k1, k2, k3) values (5, 'bar', 5)");
assertRowCount(execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE '%%ar%%' ALLOW FILTERING"), 1);
assertRowCount(execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE '%%ca%%' ALLOW FILTERING"), 0);
assertThatThrownBy(() -> execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE '%%ca%%'"))
.hasMessage(StatementRestrictions.REQUIRES_ALLOW_FILTERING_MESSAGE)
.isInstanceOf(InvalidRequestException.class);
}
@Test
public void testAllowFilteringWithLikeMatchesPostFiltering()
{
createTable("CREATE TABLE %S (k1 int, k2 text, k3 int, PRIMARY KEY (k1))");
createIndex("CREATE INDEX ON %s(k3) USING 'sai'");
execute("insert into %s (k1, k2, k3) values (1, 'fo', 1)");
execute("insert into %s (k1, k2, k3) values (2, 'foo', 2)");
execute("insert into %s (k1, k2, k3) values (3, 'fo', 3)");
execute("insert into %s (k1, k2, k3) values (4, 'ba', 4)");
execute("insert into %s (k1, k2, k3) values (5, 'bar', 5)");
assertRowCount(execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE 'foo' ALLOW FILTERING"), 1);
assertRowCount(execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE 'baar' ALLOW FILTERING"), 0);
assertThatThrownBy(() -> execute("SELECT * FROM %s WHERE k3 > 0 AND k2 LIKE 'baar'"))
.hasMessage(StatementRestrictions.REQUIRES_ALLOW_FILTERING_MESSAGE)
.isInstanceOf(InvalidRequestException.class);
}
}

View File

@ -63,7 +63,7 @@ public class UnindexedExpressionsTest extends SAITester
execute("INSERT INTO %s (pk, val1, val2) VALUES (4, 4, '44')");
// The LIKE operator is rejected because it needs to be handled by an index
assertInvalidMessage("LIKE restriction is only supported on properly indexed columns",
assertInvalidMessage(StatementRestrictions.REQUIRES_ALLOW_FILTERING_MESSAGE,
"SELECT pk FROM %s WHERE val1 = 1 AND val2 like '1%%'");
// The IS NOT operator is only valid on materialized views

View File

@ -1927,16 +1927,8 @@ public class SASIIndexTest
Assert.assertNotNull(results);
Assert.assertEquals(2, results.size());
try
{
executeCQL(CLUSTERING_CF_NAME_1 ,"SELECT * FROM %s.%s WHERE location LIKE '%%U' ALLOW FILTERING");
Assert.fail();
}
catch (InvalidRequestException e)
{
Assert.assertTrue(e.getMessage().contains("only supported"));
// expected
}
results = executeCQL(CLUSTERING_CF_NAME_1 ,"SELECT * FROM %s.%s WHERE location LIKE '%%U' ALLOW FILTERING");
Assert.assertNotNull(results);
try
{