Examples with RowSetReader org.apache.drill.exec.physical.rowSet.RowSetReader used on opensource projects

Example 26 with RowSetReader

use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.

the class TestResultSetLoaderOverflow method testVectorSizeLimitWithAppend.

@Test
public void testVectorSizeLimitWithAppend() {
    TupleMetadata schema = new SchemaBuilder().add("s", MinorType.VARCHAR).buildSchema();
    ResultSetOptions options = new ResultSetOptionBuilder().rowCountLimit(ValueVector.MAX_ROW_COUNT).readerSchema(schema).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    RowSetLoader rootWriter = rsLoader.writer();
    rsLoader.startBatch();
    byte[] head = "abc".getBytes();
    byte[] tail = new byte[523];
    Arrays.fill(tail, (byte) 'X');
    String expected = new String(head, Charsets.UTF_8);
    expected += new String(tail, Charsets.UTF_8);
    expected += new String(tail, Charsets.UTF_8);
    int count = 0;
    ScalarWriter colWriter = rootWriter.scalar(0);
    while (!rootWriter.isFull()) {
        rootWriter.start();
        colWriter.setBytes(head, head.length);
        colWriter.appendBytes(tail, tail.length);
        colWriter.appendBytes(tail, tail.length);
        rootWriter.save();
        count++;
    }
    // Number of rows should be driven by vector size.
    // Our row count should include the overflow row
    int valueLength = head.length + 2 * tail.length;
    int expectedCount = ValueVector.MAX_BUFFER_SIZE / valueLength;
    assertEquals(expectedCount + 1, count);
    // Loader's row count should include only "visible" rows
    assertEquals(expectedCount, rootWriter.rowCount());
    // Total count should include invisible and look-ahead rows.
    assertEquals(expectedCount + 1, rsLoader.totalRowCount());
    // Result should exclude the overflow row
    {
        VectorContainer container = rsLoader.harvest();
        BatchValidator.validate(container);
        RowSet result = fixture.wrap(container);
        assertEquals(expectedCount, result.rowCount());
        // Verify that the values were, in fact, appended.
        RowSetReader reader = result.reader();
        while (reader.next()) {
            assertEquals(expected, reader.scalar(0).getString());
        }
        result.clear();
    }
    // Next batch should start with the overflow row
    rsLoader.startBatch();
    assertEquals(1, rootWriter.rowCount());
    assertEquals(expectedCount + 1, rsLoader.totalRowCount());
    {
        VectorContainer container = rsLoader.harvest();
        BatchValidator.validate(container);
        RowSet result = fixture.wrap(container);
        assertEquals(1, result.rowCount());
        RowSetReader reader = result.reader();
        while (reader.next()) {
            assertEquals(expected, reader.scalar(0).getString());
        }
        result.clear();
    }
    rsLoader.close();
}

Example 27 with RowSetReader

use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.

the class TestResultSetLoaderOverflow method testArrayOverflowWithOtherArrays.

/**
 * Test the complete set of array overflow cases:
 * <ul>
 * <li>Array a is written before the column that has overflow,
 * and must be copied, in its entirety, to the overflow row.</li>
 * <li>Column b causes the overflow.</li>
 * <li>Column c is written after the overflow, and should go
 * to the look-ahead row.</li>
 * <li>Column d is written for a while, then has empties before
 * the overflow row, but is written in the overflow row.<li>
 * <li>Column e is like d, but is not written in the overflow
 * row.</li>
 */
@Test
public void testArrayOverflowWithOtherArrays() {
    TupleMetadata schema = new SchemaBuilder().addArray("a", MinorType.INT).addArray("b", MinorType.VARCHAR).addArray("c", MinorType.INT).addArray("d", MinorType.INT).buildSchema();
    ResultSetOptions options = new ResultSetOptionBuilder().rowCountLimit(ValueVector.MAX_ROW_COUNT).readerSchema(schema).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    RowSetLoader rootWriter = rsLoader.writer();
    // Fill batch with rows of with a single array, three values each. Tack on
    // a suffix to each so we can be sure the proper data is written and moved
    // to the overflow batch.
    byte[] value = new byte[512];
    Arrays.fill(value, (byte) 'X');
    String strValue = new String(value, Charsets.UTF_8);
    int aCount = 3;
    int bCount = 11;
    int cCount = 5;
    int dCount = 7;
    int cCutoff = ValueVector.MAX_BUFFER_SIZE / value.length / bCount / 2;
    ScalarWriter aWriter = rootWriter.array("a").scalar();
    ScalarWriter bWriter = rootWriter.array("b").scalar();
    ScalarWriter cWriter = rootWriter.array("c").scalar();
    ScalarWriter dWriter = rootWriter.array("d").scalar();
    int count = 0;
    rsLoader.startBatch();
    while (rootWriter.start()) {
        for (int i = 0; i < aCount; i++) {
            aWriter.setInt(count * aCount + i);
        }
        for (int i = 0; i < bCount; i++) {
            String cellValue = strValue + (count * bCount + i);
            bWriter.setString(cellValue);
        }
        if (count < cCutoff) {
            for (int i = 0; i < cCount; i++) {
                cWriter.setInt(count * cCount + i);
            }
        }
        if (count < cCutoff || rootWriter.isFull()) {
            for (int i = 0; i < dCount; i++) {
                dWriter.setInt(count * dCount + i);
            }
        }
        rootWriter.save();
        count++;
    }
    // Verify
    {
        VectorContainer container = rsLoader.harvest();
        BatchValidator.validate(container);
        RowSet result = fixture.wrap(container);
        assertEquals(count - 1, result.rowCount());
        RowSetReader reader = result.reader();
        ArrayReader aArray = reader.array("a");
        ScalarReader aReader = aArray.scalar();
        ArrayReader bArray = reader.array("b");
        ScalarReader bReader = bArray.scalar();
        ArrayReader cArray = reader.array("c");
        ScalarReader cReader = cArray.scalar();
        ArrayReader dArray = reader.array("d");
        ScalarReader dReader = dArray.scalar();
        while (reader.next()) {
            int rowId = reader.offset();
            assertEquals(aCount, aArray.size());
            for (int i = 0; i < aCount; i++) {
                assertTrue(aArray.next());
                assertEquals(rowId * aCount + i, aReader.getInt());
            }
            assertEquals(bCount, bArray.size());
            for (int i = 0; i < bCount; i++) {
                assertTrue(bArray.next());
                String cellValue = strValue + (rowId * bCount + i);
                assertEquals(cellValue, bReader.getString());
            }
            if (rowId < cCutoff) {
                assertEquals(cCount, cArray.size());
                for (int i = 0; i < cCount; i++) {
                    assertTrue(cArray.next());
                    assertEquals(rowId * cCount + i, cReader.getInt());
                }
                assertEquals(dCount, dArray.size());
                for (int i = 0; i < dCount; i++) {
                    assertTrue(dArray.next());
                    assertEquals(rowId * dCount + i, dReader.getInt());
                }
            } else {
                assertEquals(0, cArray.size());
                assertEquals(0, dArray.size());
            }
        }
        result.clear();
    }
    int firstCount = count - 1;
    // One row is in the batch. Write more, skipping over the
    // initial few values for columns c and d. Column d has a
    // roll-over value, c has an empty roll-over.
    rsLoader.startBatch();
    for (int j = 0; j < 5; j++) {
        rootWriter.start();
        for (int i = 0; i < aCount; i++) {
            aWriter.setInt(count * aCount + i);
        }
        for (int i = 0; i < bCount; i++) {
            String cellValue = strValue + (count * bCount + i);
            bWriter.setString(cellValue);
        }
        if (j > 3) {
            for (int i = 0; i < cCount; i++) {
                cWriter.setInt(count * cCount + i);
            }
            for (int i = 0; i < dCount; i++) {
                dWriter.setInt(count * dCount + i);
            }
        }
        rootWriter.save();
        count++;
    }
    {
        VectorContainer container = rsLoader.harvest();
        BatchValidator.validate(container);
        RowSet result = fixture.wrap(container);
        assertEquals(6, result.rowCount());
        RowSetReader reader = result.reader();
        ArrayReader aArray = reader.array("a");
        ScalarReader aReader = aArray.scalar();
        ArrayReader bArray = reader.array("b");
        ScalarReader bReader = bArray.scalar();
        ArrayReader cArray = reader.array("c");
        ScalarReader cReader = cArray.scalar();
        ArrayReader dArray = reader.array("d");
        ScalarReader dReader = dArray.scalar();
        int j = 0;
        while (reader.next()) {
            int rowId = firstCount + reader.offset();
            assertEquals(aCount, aArray.size());
            for (int i = 0; i < aCount; i++) {
                assertTrue(aArray.next());
                assertEquals("Index " + i, rowId * aCount + i, aReader.getInt());
            }
            assertEquals(bCount, bArray.size());
            for (int i = 0; i < bCount; i++) {
                assertTrue(bArray.next());
                String cellValue = strValue + (rowId * bCount + i);
                assertEquals(cellValue, bReader.getString());
            }
            if (j > 4) {
                assertEquals(cCount, cArray.size());
                for (int i = 0; i < cCount; i++) {
                    assertTrue(cArray.next());
                    assertEquals(rowId * cCount + i, cReader.getInt());
                }
            } else {
                assertEquals(0, cArray.size());
            }
            if (j == 0 || j > 4) {
                assertEquals(dCount, dArray.size());
                for (int i = 0; i < dCount; i++) {
                    assertTrue(dArray.next());
                    assertEquals(rowId * dCount + i, dReader.getInt());
                }
            } else {
                assertEquals(0, dArray.size());
            }
            j++;
        }
        result.clear();
    }
    rsLoader.close();
}

Example 28 with RowSetReader

use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.

the class TestResultSetLoaderUnions method testUnionOverflow.

@Test
public void testUnionOverflow() {
    final TupleMetadata schema = new SchemaBuilder().add("id", MinorType.INT).addUnion("u").addType(MinorType.INT).addType(MinorType.VARCHAR).resumeSchema().buildSchema();
    final ResultSetLoaderImpl.ResultSetOptions options = new ResultSetOptionBuilder().rowCountLimit(ValueVector.MAX_ROW_COUNT).readerSchema(schema).build();
    final ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    final RowSetLoader writer = rsLoader.writer();
    // Fill the batch with enough data to cause overflow.
    // Fill even rows with a Varchar, odd rows with an int.
    // Data must be large enough to cause overflow before 32K rows
    // (the half that get strings.
    // 16 MB / 32 K = 512 bytes
    // Make a bit bigger to overflow early.
    final int strLength = 600;
    final byte[] value = new byte[strLength - 6];
    Arrays.fill(value, (byte) 'X');
    final String strValue = new String(value, Charsets.UTF_8);
    int count = 0;
    rsLoader.startBatch();
    while (!writer.isFull()) {
        if (count % 2 == 0) {
            writer.addRow(count, String.format("%s%06d", strValue, count));
        } else {
            writer.addRow(count, count * 10);
        }
        count++;
    }
    // Number of rows should be driven by vector size.
    // Our row count should include the overflow row
    final int expectedCount = ValueVector.MAX_BUFFER_SIZE / strLength * 2;
    assertEquals(expectedCount + 1, count);
    // Loader's row count should include only "visible" rows
    assertEquals(expectedCount, writer.rowCount());
    // Total count should include invisible and look-ahead rows.
    assertEquals(expectedCount + 1, rsLoader.totalRowCount());
    // Result should exclude the overflow row
    RowSet result = fixture.wrap(rsLoader.harvest());
    assertEquals(expectedCount, result.rowCount());
    // Verify the data.
    RowSetReader reader = result.reader();
    int readCount = 0;
    while (reader.next()) {
        assertEquals(readCount, reader.scalar(0).getInt());
        if (readCount % 2 == 0) {
            assertEquals(String.format("%s%06d", strValue, readCount), reader.variant(1).scalar().getString());
        } else {
            assertEquals(readCount * 10, reader.variant(1).scalar().getInt());
        }
        readCount++;
    }
    assertEquals(readCount, result.rowCount());
    result.clear();
    // Write a few more rows to verify the overflow row.
    rsLoader.startBatch();
    for (int i = 0; i < 1000; i++) {
        if (count % 2 == 0) {
            writer.addRow(count, String.format("%s%06d", strValue, count));
        } else {
            writer.addRow(count, count * 10);
        }
        count++;
    }
    result = fixture.wrap(rsLoader.harvest());
    assertEquals(1001, result.rowCount());
    final int startCount = readCount;
    reader = result.reader();
    while (reader.next()) {
        assertEquals(readCount, reader.scalar(0).getInt());
        if (readCount % 2 == 0) {
            assertEquals(String.format("%s%06d", strValue, readCount), reader.variant(1).scalar().getString());
        } else {
            assertEquals(readCount * 10, reader.variant(1).scalar().getInt());
        }
        readCount++;
    }
    assertEquals(readCount - startCount, result.rowCount());
    result.clear();
    rsLoader.close();
}

Example 29 with RowSetReader

use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.

the class TestResultSetSchemaChange method testSchemaChangeWithOverflow.

/**
 * Test a schema change on the row that overflows. If the
 * new column is added after overflow, it will appear as
 * a schema-change in the following batch. This is fine as
 * we are essentially time-shifting: pretending that the
 * overflow row was written in the next batch (which, in
 * fact, it is: that's what overflow means.)
 */
@Test
public void testSchemaChangeWithOverflow() {
    ResultSetOptions options = new ResultSetOptionBuilder().rowCountLimit(ValueVector.MAX_ROW_COUNT).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    RowSetLoader rootWriter = rsLoader.writer();
    rootWriter.addColumn(SchemaBuilder.columnSchema("a", MinorType.VARCHAR, DataMode.REQUIRED));
    rsLoader.startBatch();
    byte[] value = new byte[512];
    Arrays.fill(value, (byte) 'X');
    int count = 0;
    while (!rootWriter.isFull()) {
        rootWriter.start();
        rootWriter.scalar(0).setBytes(value, value.length);
        if (rootWriter.isFull()) {
            rootWriter.addColumn(SchemaBuilder.columnSchema("b", MinorType.INT, DataMode.OPTIONAL));
            rootWriter.scalar(1).setInt(count);
            // Add a Varchar to ensure its offset fiddling is done properly
            rootWriter.addColumn(SchemaBuilder.columnSchema("c", MinorType.VARCHAR, DataMode.OPTIONAL));
            rootWriter.scalar(2).setString("c-" + count);
            // Allow adding a required column at this point.
            // (Not intuitively obvious that this should work; we back-fill
            // with zeros.)
            rootWriter.addColumn(SchemaBuilder.columnSchema("d", MinorType.INT, DataMode.REQUIRED));
        }
        rootWriter.save();
        count++;
    }
    // Result should include only the first column.
    SchemaBuilder schemaBuilder = new SchemaBuilder().add("a", MinorType.VARCHAR);
    BatchSchema expectedSchema = new BatchSchemaBuilder().withSchemaBuilder(schemaBuilder).build();
    RowSet result = fixture.wrap(rsLoader.harvest());
    assertTrue(result.batchSchema().isEquivalent(expectedSchema));
    assertEquals(count - 1, result.rowCount());
    result.clear();
    assertEquals(1, rsLoader.schemaVersion());
    // Double check: still can add a required column after
    // starting the next batch. (No longer in overflow state.)
    rsLoader.startBatch();
    rootWriter.addColumn(SchemaBuilder.columnSchema("e", MinorType.INT, DataMode.REQUIRED));
    // Next batch should start with the overflow row, including
    // the column added at the end of the previous batch, after
    // overflow.
    result = fixture.wrap(rsLoader.harvest());
    assertEquals(5, rsLoader.schemaVersion());
    assertEquals(1, result.rowCount());
    BatchSchemaBuilder batchSchemaBuilder = new BatchSchemaBuilder(expectedSchema);
    batchSchemaBuilder.schemaBuilder().addNullable("b", MinorType.INT).addNullable("c", MinorType.VARCHAR).add("d", MinorType.INT).add("e", MinorType.INT);
    expectedSchema = batchSchemaBuilder.build();
    assertTrue(result.batchSchema().isEquivalent(expectedSchema));
    RowSetReader reader = result.reader();
    reader.next();
    assertEquals(count - 1, reader.scalar(1).getInt());
    assertEquals("c-" + (count - 1), reader.scalar(2).getString());
    assertEquals(0, reader.scalar("d").getInt());
    assertEquals(0, reader.scalar("e").getInt());
    result.clear();
    rsLoader.close();
}

Example 30 with RowSetReader

use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.

the class TestAnalyze method verifyAnalyzeOutput.

// Helper function to verify output of ANALYZE statement
private void verifyAnalyzeOutput(String query, String message) throws Exception {
    DirectRowSet rowSet = queryBuilder().sql(query).rowSet();
    try {
        assertEquals(1, rowSet.rowCount());
        RowSetReader reader = rowSet.reader();
        assertEquals(2, reader.columnCount());
        while (reader.next()) {
            ObjectReader column = reader.column(1);
            assertEquals(message, column.isNull() ? null : column.getObject().toString());
        }
    } finally {
        rowSet.clear();
    }
}