Examples with SerializeWrite org.apache.hadoop.hive.serde2.fast.SerializeWrite used on opensource projects

Example 6 with SerializeWrite

use of org.apache.hadoop.hive.serde2.fast.SerializeWrite in project hive by apache.

the class TestBinarySortableFast method testBinarySortableFast.

private void testBinarySortableFast(SerdeRandomRowSource source, Object[][] rows, boolean[] columnSortOrderIsDesc, byte[] columnNullMarker, byte[] columnNotNullMarker, AbstractSerDe serde, StructObjectInspector rowOI, AbstractSerDe serde_fewer, StructObjectInspector writeRowOI, boolean ascending, PrimitiveTypeInfo[] primitiveTypeInfos, boolean useIncludeColumns, boolean doWriteFewerColumns, Random r) throws Throwable {
    int rowCount = rows.length;
    int columnCount = primitiveTypeInfos.length;
    boolean[] columnsToInclude = null;
    if (useIncludeColumns) {
        columnsToInclude = new boolean[columnCount];
        for (int i = 0; i < columnCount; i++) {
            columnsToInclude[i] = r.nextBoolean();
        }
    }
    int writeColumnCount = columnCount;
    if (doWriteFewerColumns) {
        writeColumnCount = writeRowOI.getAllStructFieldRefs().size();
    }
    BinarySortableSerializeWrite binarySortableSerializeWrite = new BinarySortableSerializeWrite(columnSortOrderIsDesc, columnNullMarker, columnNotNullMarker);
    // Try to serialize
    // One Writable per row.
    BytesWritable[] serializeWriteBytes = new BytesWritable[rowCount];
    int[][] perFieldWriteLengthsArray = new int[rowCount][];
    for (int i = 0; i < rowCount; i++) {
        Object[] row = rows[i];
        Output output = new Output();
        binarySortableSerializeWrite.set(output);
        int[] perFieldWriteLengths = new int[columnCount];
        for (int index = 0; index < writeColumnCount; index++) {
            Writable writable = (Writable) row[index];
            VerifyFast.serializeWrite(binarySortableSerializeWrite, primitiveTypeInfos[index], writable);
            perFieldWriteLengths[index] = output.getLength();
        }
        perFieldWriteLengthsArray[i] = perFieldWriteLengths;
        BytesWritable bytesWritable = new BytesWritable();
        bytesWritable.set(output.getData(), 0, output.getLength());
        serializeWriteBytes[i] = bytesWritable;
        if (i > 0) {
            int compareResult = serializeWriteBytes[i - 1].compareTo(serializeWriteBytes[i]);
            if ((compareResult < 0 && !ascending) || (compareResult > 0 && ascending)) {
                System.out.println("Test failed in " + (ascending ? "ascending" : "descending") + " order with " + (i - 1) + " and " + i);
                System.out.println("serialized data [" + (i - 1) + "] = " + TestBinarySortableSerDe.hexString(serializeWriteBytes[i - 1]));
                System.out.println("serialized data [" + i + "] = " + TestBinarySortableSerDe.hexString(serializeWriteBytes[i]));
                fail("Sort order of serialized " + (i - 1) + " and " + i + " are reversed!");
            }
        }
    }
    // Try to deserialize using DeserializeRead our Writable row objects created by SerializeWrite.
    for (int i = 0; i < rowCount; i++) {
        Object[] row = rows[i];
        BinarySortableDeserializeRead binarySortableDeserializeRead = new BinarySortableDeserializeRead(primitiveTypeInfos, /* useExternalBuffer */
        false, columnSortOrderIsDesc);
        BytesWritable bytesWritable = serializeWriteBytes[i];
        binarySortableDeserializeRead.set(bytesWritable.getBytes(), 0, bytesWritable.getLength());
        for (int index = 0; index < columnCount; index++) {
            if (useIncludeColumns && !columnsToInclude[index]) {
                binarySortableDeserializeRead.skipNextField();
            } else if (index >= writeColumnCount) {
                // Should come back a null.
                VerifyFast.verifyDeserializeRead(binarySortableDeserializeRead, primitiveTypeInfos[index], null);
            } else {
                Writable writable = (Writable) row[index];
                VerifyFast.verifyDeserializeRead(binarySortableDeserializeRead, primitiveTypeInfos[index], writable);
            }
        }
        if (writeColumnCount == columnCount) {
            TestCase.assertTrue(binarySortableDeserializeRead.isEndOfInputReached());
        }
        /*
       * Clip off one byte and expect to get an EOFException on the write field.
       */
        BinarySortableDeserializeRead binarySortableDeserializeRead2 = new BinarySortableDeserializeRead(primitiveTypeInfos, /* useExternalBuffer */
        false, columnSortOrderIsDesc);
        binarySortableDeserializeRead2.set(bytesWritable.getBytes(), 0, // One fewer byte.
        bytesWritable.getLength() - 1);
        for (int index = 0; index < writeColumnCount; index++) {
            Writable writable = (Writable) row[index];
            if (index == writeColumnCount - 1) {
                boolean threw = false;
                try {
                    VerifyFast.verifyDeserializeRead(binarySortableDeserializeRead2, primitiveTypeInfos[index], writable);
                } catch (EOFException e) {
                    //          debugDetailedReadPositionString = binarySortableDeserializeRead2.getDetailedReadPositionString();
                    //          debugStackTrace = e.getStackTrace();
                    threw = true;
                }
                TestCase.assertTrue(threw);
            } else {
                if (useIncludeColumns && !columnsToInclude[index]) {
                    binarySortableDeserializeRead2.skipNextField();
                } else {
                    VerifyFast.verifyDeserializeRead(binarySortableDeserializeRead2, primitiveTypeInfos[index], writable);
                }
            }
        }
    }
    // Try to deserialize using SerDe class our Writable row objects created by SerializeWrite.
    for (int i = 0; i < rowCount; i++) {
        BytesWritable bytesWritable = serializeWriteBytes[i];
        // Note that regular SerDe doesn't tolerate fewer columns.
        List<Object> deserializedRow;
        if (doWriteFewerColumns) {
            deserializedRow = (List<Object>) serde_fewer.deserialize(bytesWritable);
        } else {
            deserializedRow = (List<Object>) serde.deserialize(bytesWritable);
        }
        Object[] row = rows[i];
        for (int index = 0; index < writeColumnCount; index++) {
            Object expected = row[index];
            Object object = deserializedRow.get(index);
            if (expected == null || object == null) {
                if (expected != null || object != null) {
                    fail("SerDe deserialized NULL column mismatch");
                }
            } else {
                if (!object.equals(expected)) {
                    fail("SerDe deserialized value does not match (expected " + expected.getClass().getName() + " " + expected.toString() + ", actual " + object.getClass().getName() + " " + object.toString() + ")");
                }
            }
        }
    }
    // One Writable per row.
    BytesWritable[] serdeBytes = new BytesWritable[rowCount];
    // Serialize using the SerDe, then below deserialize using DeserializeRead.
    for (int i = 0; i < rowCount; i++) {
        Object[] row = rows[i];
        // Since SerDe reuses memory, we will need to make a copy.
        BytesWritable serialized;
        if (doWriteFewerColumns) {
            serialized = (BytesWritable) serde_fewer.serialize(row, rowOI);
        } else {
            serialized = (BytesWritable) serde.serialize(row, rowOI);
            ;
        }
        BytesWritable bytesWritable = new BytesWritable();
        bytesWritable.set(serialized);
        byte[] serDeOutput = Arrays.copyOfRange(bytesWritable.getBytes(), 0, bytesWritable.getLength());
        byte[] serializeWriteExpected = Arrays.copyOfRange(serializeWriteBytes[i].getBytes(), 0, serializeWriteBytes[i].getLength());
        if (!Arrays.equals(serDeOutput, serializeWriteExpected)) {
            int mismatchPos = -1;
            if (serDeOutput.length != serializeWriteExpected.length) {
                for (int b = 0; b < Math.min(serDeOutput.length, serializeWriteExpected.length); b++) {
                    if (serDeOutput[b] != serializeWriteExpected[b]) {
                        mismatchPos = b;
                        break;
                    }
                }
                fail("Different byte array lengths: serDeOutput.length " + serDeOutput.length + ", serializeWriteExpected.length " + serializeWriteExpected.length + " mismatchPos " + mismatchPos + " perFieldWriteLengths " + Arrays.toString(perFieldWriteLengthsArray[i]));
            }
            List<Integer> differentPositions = new ArrayList();
            for (int b = 0; b < serDeOutput.length; b++) {
                if (serDeOutput[b] != serializeWriteExpected[b]) {
                    differentPositions.add(b);
                }
            }
            if (differentPositions.size() > 0) {
                List<String> serializeWriteExpectedFields = new ArrayList<String>();
                List<String> serDeFields = new ArrayList<String>();
                int f = 0;
                int lastBegin = 0;
                for (int b = 0; b < serDeOutput.length; b++) {
                    int writeLength = perFieldWriteLengthsArray[i][f];
                    if (b + 1 == writeLength) {
                        serializeWriteExpectedFields.add(displayBytes(serializeWriteExpected, lastBegin, writeLength - lastBegin));
                        serDeFields.add(displayBytes(serDeOutput, lastBegin, writeLength - lastBegin));
                        f++;
                        lastBegin = b + 1;
                    }
                }
                fail("SerializeWrite and SerDe serialization does not match at positions " + differentPositions.toString() + "\n(SerializeWrite: " + serializeWriteExpectedFields.toString() + "\nSerDe: " + serDeFields.toString() + "\nperFieldWriteLengths " + Arrays.toString(perFieldWriteLengthsArray[i]) + "\nprimitiveTypeInfos " + Arrays.toString(primitiveTypeInfos) + "\nrow " + Arrays.toString(row));
            }
        }
        serdeBytes[i] = bytesWritable;
    }
    // Try to deserialize using DeserializeRead our Writable row objects created by SerDe.
    for (int i = 0; i < rowCount; i++) {
        Object[] row = rows[i];
        BinarySortableDeserializeRead binarySortableDeserializeRead = new BinarySortableDeserializeRead(primitiveTypeInfos, /* useExternalBuffer */
        false, columnSortOrderIsDesc);
        BytesWritable bytesWritable = serdeBytes[i];
        binarySortableDeserializeRead.set(bytesWritable.getBytes(), 0, bytesWritable.getLength());
        for (int index = 0; index < columnCount; index++) {
            if (useIncludeColumns && !columnsToInclude[index]) {
                binarySortableDeserializeRead.skipNextField();
            } else if (index >= writeColumnCount) {
                // Should come back a null.
                VerifyFast.verifyDeserializeRead(binarySortableDeserializeRead, primitiveTypeInfos[index], null);
            } else {
                Writable writable = (Writable) row[index];
                VerifyFast.verifyDeserializeRead(binarySortableDeserializeRead, primitiveTypeInfos[index], writable);
            }
        }
        if (writeColumnCount == columnCount) {
            TestCase.assertTrue(binarySortableDeserializeRead.isEndOfInputReached());
        }
    }
}

Example 7 with SerializeWrite

use of org.apache.hadoop.hive.serde2.fast.SerializeWrite in project hive by apache.

the class TestVectorSerDeRow method testVectorDeserializeRow.

void testVectorDeserializeRow(Random r, SerializationType serializationType, boolean alternate1, boolean alternate2, boolean useExternalBuffer) throws HiveException, IOException, SerDeException {
    String[] emptyScratchTypeNames = new String[0];
    VectorRandomRowSource source = new VectorRandomRowSource();
    source.init(r);
    VectorizedRowBatchCtx batchContext = new VectorizedRowBatchCtx();
    batchContext.init(source.rowStructObjectInspector(), emptyScratchTypeNames);
    VectorizedRowBatch batch = batchContext.createVectorizedRowBatch();
    // junk the destination for the 1st pass
    for (ColumnVector cv : batch.cols) {
        Arrays.fill(cv.isNull, true);
    }
    PrimitiveTypeInfo[] primitiveTypeInfos = source.primitiveTypeInfos();
    int fieldCount = source.typeNames().size();
    DeserializeRead deserializeRead;
    SerializeWrite serializeWrite;
    switch(serializationType) {
        case BINARY_SORTABLE:
            boolean useColumnSortOrderIsDesc = alternate1;
            if (!useColumnSortOrderIsDesc) {
                deserializeRead = new BinarySortableDeserializeRead(source.primitiveTypeInfos(), useExternalBuffer);
                serializeWrite = new BinarySortableSerializeWrite(fieldCount);
            } else {
                boolean[] columnSortOrderIsDesc = new boolean[fieldCount];
                for (int i = 0; i < fieldCount; i++) {
                    columnSortOrderIsDesc[i] = r.nextBoolean();
                }
                deserializeRead = new BinarySortableDeserializeRead(source.primitiveTypeInfos(), useExternalBuffer, columnSortOrderIsDesc);
                byte[] columnNullMarker = new byte[fieldCount];
                byte[] columnNotNullMarker = new byte[fieldCount];
                for (int i = 0; i < fieldCount; i++) {
                    if (columnSortOrderIsDesc[i]) {
                        // Descending
                        // Null last (default for descending order)
                        columnNullMarker[i] = BinarySortableSerDe.ZERO;
                        columnNotNullMarker[i] = BinarySortableSerDe.ONE;
                    } else {
                        // Ascending
                        // Null first (default for ascending order)
                        columnNullMarker[i] = BinarySortableSerDe.ZERO;
                        columnNotNullMarker[i] = BinarySortableSerDe.ONE;
                    }
                }
                serializeWrite = new BinarySortableSerializeWrite(columnSortOrderIsDesc, columnNullMarker, columnNotNullMarker);
            }
            boolean useBinarySortableCharsNeedingEscape = alternate2;
            if (useBinarySortableCharsNeedingEscape) {
                source.addBinarySortableAlphabets();
            }
            break;
        case LAZY_BINARY:
            deserializeRead = new LazyBinaryDeserializeRead(source.primitiveTypeInfos(), useExternalBuffer);
            serializeWrite = new LazyBinarySerializeWrite(fieldCount);
            break;
        case LAZY_SIMPLE:
            {
                StructObjectInspector rowObjectInspector = source.rowStructObjectInspector();
                Configuration conf = new Configuration();
                Properties tbl = new Properties();
                tbl.setProperty(serdeConstants.FIELD_DELIM, "\t");
                tbl.setProperty(serdeConstants.LINE_DELIM, "\n");
                byte separator = (byte) '\t';
                boolean useLazySimpleEscapes = alternate1;
                if (useLazySimpleEscapes) {
                    tbl.setProperty(serdeConstants.QUOTE_CHAR, "'");
                    String escapeString = "\\";
                    tbl.setProperty(serdeConstants.ESCAPE_CHAR, escapeString);
                }
                LazySerDeParameters lazySerDeParams = getSerDeParams(conf, tbl, rowObjectInspector);
                if (useLazySimpleEscapes) {
                    // LazySimple seems to throw away everything but \n and \r.
                    boolean[] needsEscape = lazySerDeParams.getNeedsEscape();
                    StringBuilder sb = new StringBuilder();
                    if (needsEscape['\n']) {
                        sb.append('\n');
                    }
                    if (needsEscape['\r']) {
                        sb.append('\r');
                    }
                    // for (int i = 0; i < needsEscape.length; i++) {
                    //  if (needsEscape[i]) {
                    //    sb.append((char) i);
                    //  }
                    // }
                    String needsEscapeStr = sb.toString();
                    if (needsEscapeStr.length() > 0) {
                        source.addEscapables(needsEscapeStr);
                    }
                }
                deserializeRead = new LazySimpleDeserializeRead(source.primitiveTypeInfos(), useExternalBuffer, separator, lazySerDeParams);
                serializeWrite = new LazySimpleSerializeWrite(fieldCount, separator, lazySerDeParams);
            }
            break;
        default:
            throw new Error("Unknown serialization type " + serializationType);
    }
    VectorDeserializeRow vectorDeserializeRow = new VectorDeserializeRow(deserializeRead);
    vectorDeserializeRow.init();
    // junk the destination for the 1st pass
    for (ColumnVector cv : batch.cols) {
        Arrays.fill(cv.isNull, true);
        cv.noNulls = false;
    }
    VectorExtractRow vectorExtractRow = new VectorExtractRow();
    vectorExtractRow.init(source.typeNames());
    Object[][] randomRows = source.randomRows(100000);
    int firstRandomRowIndex = 0;
    for (int i = 0; i < randomRows.length; i++) {
        Object[] row = randomRows[i];
        Output output = serializeRow(row, source, serializeWrite);
        vectorDeserializeRow.setBytes(output.getData(), 0, output.getLength());
        try {
            vectorDeserializeRow.deserialize(batch, batch.size);
        } catch (Exception e) {
            throw new HiveException("\nDeserializeRead details: " + vectorDeserializeRow.getDetailedReadPositionString(), e);
        }
        batch.size++;
        if (batch.size == batch.DEFAULT_SIZE) {
            examineBatch(batch, vectorExtractRow, primitiveTypeInfos, randomRows, firstRandomRowIndex);
            firstRandomRowIndex = i + 1;
            batch.reset();
        }
    }
    if (batch.size > 0) {
        examineBatch(batch, vectorExtractRow, primitiveTypeInfos, randomRows, firstRandomRowIndex);
    }
}