Examples with REPLICATED io.trino.sql.planner.plan.SemiJoinNode.DistributionType.REPLICATED used on opensource projects

Example 1 with REPLICATED

use of io.trino.sql.planner.plan.SemiJoinNode.DistributionType.REPLICATED in project trino by trinodb.

the class TestDetermineSemiJoinDistributionType method testReplicatesWhenNotRestricted.

@Test
public void testReplicatesWhenNotRestricted() {
    // variable width so that average row size is respected
    Type symbolType = createUnboundedVarcharType();
    int aRows = 10_000;
    int bRows = 10;
    PlanNodeStatsEstimate probeSideStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(aRows).addSymbolStatistics(ImmutableMap.of(new Symbol("A1"), new SymbolStatsEstimate(0, 100, 0, 640000, 10))).build();
    PlanNodeStatsEstimate buildSideStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(bRows).addSymbolStatistics(ImmutableMap.of(new Symbol("B1"), new SymbolStatsEstimate(0, 100, 0, 640000, 10))).build();
    // B table is small enough to be replicated in AUTOMATIC_RESTRICTED mode
    assertDetermineSemiJoinDistributionType().setSystemProperty(JOIN_DISTRIBUTION_TYPE, JoinDistributionType.AUTOMATIC.name()).setSystemProperty(JOIN_MAX_BROADCAST_TABLE_SIZE, "100MB").overrideStats("valuesA", probeSideStatsEstimate).overrideStats("valuesB", buildSideStatsEstimate).on(p -> {
        Symbol a1 = p.symbol("A1", symbolType);
        Symbol b1 = p.symbol("B1", symbolType);
        return p.semiJoin(p.values(new PlanNodeId("valuesA"), aRows, a1), p.values(new PlanNodeId("valuesB"), bRows, b1), a1, b1, p.symbol("output"), Optional.empty(), Optional.empty(), Optional.empty());
    }).matches(semiJoin("A1", "B1", "output", Optional.of(REPLICATED), values(ImmutableMap.of("A1", 0)), values(ImmutableMap.of("B1", 0))));
    probeSideStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(aRows).addSymbolStatistics(ImmutableMap.of(new Symbol("A1"), new SymbolStatsEstimate(0, 100, 0, 640000d * 10000, 10))).build();
    buildSideStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(bRows).addSymbolStatistics(ImmutableMap.of(new Symbol("B1"), new SymbolStatsEstimate(0, 100, 0, 640000d * 10000, 10))).build();
    // B table exceeds AUTOMATIC_RESTRICTED limit therefore it is partitioned
    assertDetermineSemiJoinDistributionType().setSystemProperty(JOIN_DISTRIBUTION_TYPE, JoinDistributionType.AUTOMATIC.name()).setSystemProperty(JOIN_MAX_BROADCAST_TABLE_SIZE, "100MB").overrideStats("valuesA", probeSideStatsEstimate).overrideStats("valuesB", buildSideStatsEstimate).on(p -> {
        Symbol a1 = p.symbol("A1", symbolType);
        Symbol b1 = p.symbol("B1", symbolType);
        return p.semiJoin(p.values(new PlanNodeId("valuesA"), aRows, a1), p.values(new PlanNodeId("valuesB"), bRows, b1), a1, b1, p.symbol("output"), Optional.empty(), Optional.empty(), Optional.empty());
    }).matches(semiJoin("A1", "B1", "output", Optional.of(PARTITIONED), values(ImmutableMap.of("A1", 0)), values(ImmutableMap.of("B1", 0))));
}

Example 2 with REPLICATED

use of io.trino.sql.planner.plan.SemiJoinNode.DistributionType.REPLICATED in project trino by trinodb.

the class TestDetermineSemiJoinDistributionType method testReplicatesWhenSourceIsSmall.

@Test
public void testReplicatesWhenSourceIsSmall() {
    // variable width so that average row size is respected
    Type symbolType = createUnboundedVarcharType();
    int aRows = 10_000;
    int bRows = 10;
    PlanNodeStatsEstimate probeSideStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(aRows).addSymbolStatistics(ImmutableMap.of(new Symbol("A1"), new SymbolStatsEstimate(0, 100, 0, 640000d * 10000, 10))).build();
    PlanNodeStatsEstimate buildSideStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(bRows).addSymbolStatistics(ImmutableMap.of(new Symbol("B1"), new SymbolStatsEstimate(0, 100, 0, 640000d * 10000, 10))).build();
    PlanNodeStatsEstimate buildSideSourceStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(bRows).addSymbolStatistics(ImmutableMap.of(new Symbol("B1"), new SymbolStatsEstimate(0, 100, 0, 64, 10))).build();
    // build side exceeds AUTOMATIC_RESTRICTED limit but source plan nodes are small
    // therefore replicated distribution type is chosen
    assertDetermineSemiJoinDistributionType().setSystemProperty(JOIN_DISTRIBUTION_TYPE, JoinDistributionType.AUTOMATIC.name()).setSystemProperty(JOIN_MAX_BROADCAST_TABLE_SIZE, "100MB").overrideStats("valuesA", probeSideStatsEstimate).overrideStats("filterB", buildSideStatsEstimate).overrideStats("valuesB", buildSideSourceStatsEstimate).on(p -> {
        Symbol a1 = p.symbol("A1", symbolType);
        Symbol b1 = p.symbol("B1", symbolType);
        return p.semiJoin(p.values(new PlanNodeId("valuesA"), aRows, a1), p.filter(new PlanNodeId("filterB"), TRUE_LITERAL, p.values(new PlanNodeId("valuesB"), bRows, b1)), a1, b1, p.symbol("output"), Optional.empty(), Optional.empty(), Optional.empty());
    }).matches(semiJoin("A1", "B1", "output", Optional.of(REPLICATED), values(ImmutableMap.of("A1", 0)), filter("true", values(ImmutableMap.of("B1", 0)))));
}

Example 3 with REPLICATED

use of io.trino.sql.planner.plan.SemiJoinNode.DistributionType.REPLICATED in project trino by trinodb.

the class TestDetermineSemiJoinDistributionType method testReplicatesWhenRequiredBySession.

@Test
public void testReplicatesWhenRequiredBySession() {
    int aRows = 10_000;
    int bRows = 10_000;
    assertDetermineSemiJoinDistributionType().setSystemProperty(JOIN_DISTRIBUTION_TYPE, JoinDistributionType.BROADCAST.name()).setSystemProperty(JOIN_MAX_BROADCAST_TABLE_SIZE, "1B").overrideStats("valuesA", PlanNodeStatsEstimate.builder().setOutputRowCount(aRows).addSymbolStatistics(ImmutableMap.of(new Symbol("A1"), SymbolStatsEstimate.unknown())).build()).overrideStats("valuesB", PlanNodeStatsEstimate.builder().setOutputRowCount(bRows).addSymbolStatistics(ImmutableMap.of(new Symbol("B1"), SymbolStatsEstimate.unknown())).build()).on(p -> p.semiJoin(p.values(new PlanNodeId("valuesA"), aRows, p.symbol("A1", BIGINT)), p.values(new PlanNodeId("valuesB"), bRows, p.symbol("B1", BIGINT)), p.symbol("A1"), p.symbol("B1"), p.symbol("output"), Optional.empty(), Optional.empty(), Optional.empty())).matches(semiJoin("A1", "B1", "output", Optional.of(REPLICATED), values(ImmutableMap.of("A1", 0)), values(ImmutableMap.of("B1", 0))));
}

Example 4 with REPLICATED

use of io.trino.sql.planner.plan.SemiJoinNode.DistributionType.REPLICATED in project trino by trinodb.

the class TestDetermineSemiJoinDistributionType method testReplicatesWhenFilterMuchSmaller.

@Test
public void testReplicatesWhenFilterMuchSmaller() {
    int aRows = 10_000;
    int bRows = 100;
    assertDetermineSemiJoinDistributionType().setSystemProperty(JOIN_DISTRIBUTION_TYPE, JoinDistributionType.AUTOMATIC.name()).overrideStats("valuesA", PlanNodeStatsEstimate.builder().setOutputRowCount(aRows).addSymbolStatistics(ImmutableMap.of(new Symbol("A1"), SymbolStatsEstimate.unknown())).build()).overrideStats("valuesB", PlanNodeStatsEstimate.builder().setOutputRowCount(bRows).addSymbolStatistics(ImmutableMap.of(new Symbol("B1"), SymbolStatsEstimate.unknown())).build()).on(p -> p.semiJoin(p.values(new PlanNodeId("valuesA"), aRows, p.symbol("A1", BIGINT)), p.values(new PlanNodeId("valuesB"), bRows, p.symbol("B1", BIGINT)), p.symbol("A1"), p.symbol("B1"), p.symbol("output"), Optional.empty(), Optional.empty(), Optional.empty())).matches(semiJoin("A1", "B1", "output", Optional.of(REPLICATED), values(ImmutableMap.of("A1", 0)), values(ImmutableMap.of("B1", 0))));
}