SQL Physical Operator Fixes and Enhancements

Author: damccorm

sdks/java/extensions/sql/src/main/java/org/apache/beam/sdk/extensions/sql/impl/planner/BeamRuleSets.java

@@ -151,13 +151,14 @@ public class BeamRuleSets {
       ImmutableList.of(BeamEnumerableConverterRule.INSTANCE);
 
   public static Collection<RuleSet> getRuleSets() {
+    return ImmutableList.of(RuleSets.ofList(getAllRules()));
+  }
 
-    return ImmutableList.of(
-        RuleSets.ofList(
-            ImmutableList.<RelOptRule>builder()
-                .addAll(BEAM_CONVERTERS)
-                .addAll(BEAM_TO_ENUMERABLE)
-                .addAll(LOGICAL_OPTIMIZATIONS)
-                .build()));
+  public static List<RelOptRule> getAllRules() {
+    return ImmutableList.<RelOptRule>builder()
+        .addAll(BEAM_CONVERTERS)
+        .addAll(BEAM_TO_ENUMERABLE)
+        .addAll(LOGICAL_OPTIMIZATIONS)
+        .build();
   }
 }

sdks/java/extensions/sql/src/main/java/org/apache/beam/sdk/extensions/sql/impl/planner/RelMdNodeStats.java

@@ -57,6 +57,12 @@ public NodeStats getNodeStats(RelNode rel, RelMetadataQuery mq) {
       return this.getBeamNodeStats((BeamRelNode) rel, bmq);
     }
 
+    if (rel instanceof org.apache.beam.vendor.calcite.v1_40_0.org.apache.calcite.rel.core.Values) {
+      org.apache.beam.vendor.calcite.v1_40_0.org.apache.calcite.rel.core.Values values =
+          (org.apache.beam.vendor.calcite.v1_40_0.org.apache.calcite.rel.core.Values) rel;
+      return NodeStats.create(values.getTuples().size());
+    }
+
     // We can later define custom methods for all different RelNodes to prevent hitting this point.
     // Similar to RelMdRowCount in calcite.
 

sdks/java/extensions/sql/src/main/java/org/apache/beam/sdk/extensions/sql/impl/rel/BeamCalcRel.java

@@ -269,15 +269,18 @@ public CalcFn(
         List<String> jarPaths,
         FieldAccessDescriptor fieldAccess,
         boolean collectErrors) {
-      this.processElementBlock = processElementBlock;
+      this.processElementBlock =
+          processElementBlock.replace(
+              "(byte[]) org.apache.beam.vendor.calcite.v1_40_0.org.apache.calcite.runtime.SqlFunctions.concat",
+              "org.apache.beam.vendor.calcite.v1_40_0.org.apache.calcite.runtime.SqlFunctions.concat");
       this.outputSchema = outputSchema;
       this.verifyRowValues = verifyRowValues;
       this.jarPaths = jarPaths;
       this.fieldAccess = fieldAccess;
       this.collectErrors = collectErrors;
 
       // validate generated code
-      compile(processElementBlock, jarPaths);
+      compile(this.processElementBlock, jarPaths);
     }
 
     private static ScriptEvaluator compile(String processElementBlock, List<String> jarPaths) {

sdks/java/extensions/sql/src/main/java/org/apache/beam/sdk/extensions/sql/impl/rel/BeamEnumerableConverter.java

@@ -159,11 +159,28 @@ static List<Row> toRowList(PipelineOptions options, BeamRelNode node) {
     if (node instanceof BeamIOSinkRel) {
       throw new UnsupportedOperationException("Does not support BeamIOSinkRel in toRowList.");
     } else if (isLimitQuery(node)) {
-      throw new UnsupportedOperationException("Does not support queries with LIMIT in toRowList.");
+      return limitRowList(options, node);
     }
     return collectRows(options, node).stream().collect(Collectors.toList());
   }
 
+  private static List<Row> limitRowList(PipelineOptions options, BeamRelNode node) {
+    long id = options.getOptionsId();
+    ConcurrentLinkedQueue<Row> values = new ConcurrentLinkedQueue<>();
+    int limitCount = getLimitCount(node);
+
+    Collector.globalValues.put(id, values);
+    limitRun(options, node, new Collector(), values, limitCount);
+    Collector.globalValues.remove(id);
+
+    // remove extra retrieved values
+    while (values.size() > limitCount) {
+      values.remove();
+    }
+
+    return values.stream().collect(Collectors.toList());
+  }
+
   static Enumerable<Object> toEnumerable(PipelineOptions options, BeamRelNode node) {
     if (node instanceof BeamIOSinkRel) {
       return count(options, node);

sdks/java/extensions/sql/src/main/java/org/apache/beam/sdk/extensions/sql/impl/rel/BeamSetOperatorRelBase.java

@@ -17,8 +17,6 @@
  */
 package org.apache.beam.sdk.extensions.sql.impl.rel;
 
-import static org.apache.beam.vendor.guava.v32_1_2_jre.com.google.common.base.Preconditions.checkArgument;
-
 import java.io.Serializable;
 import org.apache.beam.sdk.extensions.sql.impl.transform.BeamSetOperatorsTransforms;
 import org.apache.beam.sdk.schemas.transforms.CoGroup;
@@ -59,14 +57,27 @@ public BeamSetOperatorRelBase(BeamRelNode beamRelNode, OpType opType, boolean al
 
   @Override
   public PCollection<Row> expand(PCollectionList<Row> inputs) {
-    checkArgument(
-        inputs.size() == 2,
-        "Wrong number of arguments to %s: %s",
-        beamRelNode.getClass().getSimpleName(),
-        inputs);
-    PCollection<Row> leftRows = inputs.get(0);
-    PCollection<Row> rightRows = inputs.get(1);
+    // Reverted Flatten optimization as it fails when inputs have slightly different schemas (e.g.
+    // NULL vs VARCHAR)
+    // if (opType == OpType.UNION && all) {
+    //   return inputs.apply("UnionAllFlatten", Flatten.pCollections());
+    // }
+
+    if (inputs.size() == 2) {
+      return expandPair(inputs.get(0), inputs.get(1));
+    } else if (inputs.size() > 2) {
+      PCollection<Row> result = inputs.get(0);
+      for (int i = 1; i < inputs.size(); i++) {
+        result = expandPair(result, inputs.get(i));
+      }
+      return result;
+    } else {
+      throw new IllegalArgumentException(
+          "Too few arguments to " + beamRelNode.getClass().getSimpleName());
+    }
+  }
 
+  private PCollection<Row> expandPair(PCollection<Row> leftRows, PCollection<Row> rightRows) {
     WindowFn leftWindow = leftRows.getWindowingStrategy().getWindowFn();
     WindowFn rightWindow = rightRows.getWindowingStrategy().getWindowFn();
     if (!leftWindow.isCompatible(rightWindow)) {

sdks/java/extensions/sql/src/main/java/org/apache/beam/sdk/extensions/sql/impl/rel/BeamValuesRel.java

@@ -30,8 +30,10 @@
 import org.apache.beam.sdk.extensions.sql.impl.planner.NodeStats;
 import org.apache.beam.sdk.extensions.sql.impl.utils.CalciteUtils;
 import org.apache.beam.sdk.schemas.Schema;
-import org.apache.beam.sdk.transforms.Create;
+import org.apache.beam.sdk.transforms.DoFn;
+import org.apache.beam.sdk.transforms.Impulse;
 import org.apache.beam.sdk.transforms.PTransform;
+import org.apache.beam.sdk.transforms.ParDo;
 import org.apache.beam.sdk.values.PCollection;
 import org.apache.beam.sdk.values.PCollectionList;
 import org.apache.beam.sdk.values.Row;
@@ -84,15 +86,41 @@ public PCollection<Row> expand(PCollectionList<Row> pinput) {
           BeamValuesRel.class.getSimpleName(),
           pinput);
 
-      Schema schema = CalciteUtils.toSchema(getRowType());
+      Schema inferredSchema = CalciteUtils.toSchema(getRowType());
+      Schema.Builder schemaBuilder = Schema.builder();
+      for (int i = 0; i < inferredSchema.getFieldCount(); i++) {
+        Schema.Field field = inferredSchema.getField(i);
+        boolean hasNull = false;
+        for (ImmutableList<RexLiteral> tuple : tuples) {
+          if (tuple.get(i).getValue() == null) {
+            hasNull = true;
+            break;
+          }
+        }
+        if (hasNull && !field.getType().getNullable()) {
+          schemaBuilder.addField(field.getName(), field.getType().withNullable(true));
+        } else {
+          schemaBuilder.addField(field);
+        }
+      }
+      Schema schema = schemaBuilder.build();
       List<Row> rows = tuples.stream().map(tuple -> tupleToRow(schema, tuple)).collect(toList());
-      return pinput.getPipeline().begin().apply(Create.of(rows).withRowSchema(schema));
+      return pinput
+          .getPipeline()
+          .begin()
+          .apply(Impulse.create())
+          .apply(ParDo.of(new EmitRowsFn(rows)))
+          .setRowSchema(schema);
     }
   }
 
   private Row tupleToRow(Schema schema, ImmutableList<RexLiteral> tuple) {
     return IntStream.range(0, tuple.size())
-        .mapToObj(i -> autoCastField(schema.getField(i), tuple.get(i).getValue()))
+        .mapToObj(
+            i -> {
+              Object val = tuple.get(i).getValue();
+              return autoCastField(schema.getField(i), val);
+            })
         .collect(toRow(schema));
   }
 
@@ -106,4 +134,19 @@ public BeamCostModel beamComputeSelfCost(RelOptPlanner planner, BeamRelMetadataQ
     NodeStats estimates = BeamSqlRelUtils.getNodeStats(this, mq);
     return BeamCostModel.FACTORY.makeCost(estimates.getRowCount(), estimates.getRate());
   }
+
+  private static class EmitRowsFn extends DoFn<byte[], Row> {
+    private final List<Row> rows;
+
+    public EmitRowsFn(List<Row> rows) {
+      this.rows = rows;
+    }
+
+    @ProcessElement
+    public void processElement(ProcessContext c) {
+      for (Row row : rows) {
+        c.output(row);
+      }
+    }
+  }
 }

sdks/java/extensions/sql/src/main/java/org/apache/beam/sdk/extensions/sql/impl/transform/BeamBuiltinAggregations.java

@@ -61,6 +61,11 @@ public class BeamBuiltinAggregations {
       BUILTIN_AGGREGATOR_FACTORIES =
           ImmutableMap.<String, Function<Schema.FieldType, CombineFn<?, ?, ?>>>builder()
               .put("ANY_VALUE", typeName -> Sample.anyValueCombineFn())
+              // SINGLE_VALUE is emitted by Calcite to enforce the cardinality of a scalar
+              // subquery (a subquery used as a scalar must yield exactly one row). The single
+              // input value is returned as-is; unlike COUNT/SUM it must not drop nulls, so a
+              // scalar subquery evaluating to NULL surfaces NULL.
+              .put("SINGLE_VALUE", typeName -> Sample.anyValueCombineFn())
               // Drop null elements for these aggregations.
               .put("COUNT", typeName -> new DropNullFnWithDefault(Count.combineFn()))
               .put("MAX", typeName -> new DropNullFn(BeamBuiltinAggregations.createMax(typeName)))

sdks/java/extensions/sql/src/test/java/org/apache/beam/sdk/extensions/sql/impl/rel/BeamEnumerableConverterTest.java

@@ -23,8 +23,10 @@
 
 import java.math.BigDecimal;
 import java.util.List;
+import org.apache.beam.sdk.extensions.sql.impl.BeamSqlEnv;
 import org.apache.beam.sdk.extensions.sql.impl.utils.CalciteUtils;
 import org.apache.beam.sdk.extensions.sql.meta.SchemaBaseBeamTable;
+import org.apache.beam.sdk.extensions.sql.meta.provider.test.TestBoundedTable;
 import org.apache.beam.sdk.options.PipelineOptions;
 import org.apache.beam.sdk.options.PipelineOptionsFactory;
 import org.apache.beam.sdk.schemas.Schema;
@@ -125,6 +127,21 @@ public void testToListRow_collectMultiple() {
       assertEquals(Row.withSchema(schema).addValues(0L, 1L).build(), rowList.get(0));
     }
 
+    @Test
+    public void testToRowList_limit() {
+      Schema schema = Schema.builder().addInt64Field("id").build();
+      java.util.Map<String, org.apache.beam.sdk.extensions.sql.meta.BeamSqlTable> tables =
+          new java.util.HashMap<>();
+      tables.put("TEST", TestBoundedTable.of(Schema.FieldType.INT64, "id").addRows(1L, 2L, 3L));
+      BeamSqlEnv env = BeamSqlEnv.readOnly("test", tables);
+      BeamRelNode node = env.parseQuery("SELECT id FROM TEST LIMIT 2");
+
+      List<Row> rowList = BeamEnumerableConverter.toRowList(options, node);
+      assertEquals(2, rowList.size());
+      assertTrue(rowList.contains(Row.withSchema(schema).addValue(1L).build()));
+      assertTrue(rowList.contains(Row.withSchema(schema).addValue(2L).build()));
+    }
+
     private static class FakeTable extends SchemaBaseBeamTable {
       public FakeTable() {
         super(null);