/*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.facebook.presto.hive;
import com.facebook.presto.hive.util.SerDeUtils;
import com.facebook.presto.spi.ColumnType;
import com.google.common.base.Charsets;
import com.google.common.base.Throwables;
import org.apache.hadoop.hive.metastore.api.MetaException;
import org.apache.hadoop.hive.serde2.SerDeException;
import org.apache.hadoop.hive.serde2.columnar.BytesRefArrayWritable;
import org.apache.hadoop.hive.serde2.columnar.BytesRefWritable;
import org.apache.hadoop.hive.serde2.io.TimestampWritable;
import org.apache.hadoop.hive.serde2.lazy.ByteArrayRef;
import org.apache.hadoop.hive.serde2.lazybinary.LazyBinaryFactory;
import org.apache.hadoop.hive.serde2.lazybinary.LazyBinaryObject;
import org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector;
import org.apache.hadoop.hive.serde2.objectinspector.StructObjectInspector;
import org.apache.hadoop.io.WritableUtils;
import org.apache.hadoop.mapred.RecordReader;
import sun.misc.Unsafe;
import java.io.IOException;
import java.lang.reflect.Field;
import java.util.Arrays;
import java.util.List;
import java.util.Map;
import java.util.Properties;
import java.util.Set;
import static com.facebook.presto.hive.HiveBooleanParser.isFalse;
import static com.facebook.presto.hive.HiveBooleanParser.isTrue;
import static com.facebook.presto.hive.HiveUtil.getTableObjectInspector;
import static com.facebook.presto.hive.NumberParser.parseDouble;
import static com.facebook.presto.hive.NumberParser.parseLong;
import static com.google.common.base.Preconditions.checkArgument;
import static com.google.common.base.Preconditions.checkNotNull;
import static com.google.common.base.Preconditions.checkState;
import static com.google.common.collect.Maps.uniqueIndex;
import static com.google.common.collect.Sets.immutableEnumSet;
import static java.lang.Math.max;
import static java.lang.Math.min;
class ColumnarBinaryHiveRecordCursor<K>
extends HiveRecordCursor
{
private final RecordReader<K, BytesRefArrayWritable> recordReader;
private final K key;
private final BytesRefArrayWritable value;
@SuppressWarnings("FieldCanBeLocal") // include names for debugging
private final String[] names;
private final ColumnType[] types;
private final HiveType[] hiveTypes;
private final ObjectInspector[] fieldInspectors; // DON'T USE THESE UNLESS EXTRACTION WILL BE SLOW ANYWAY
private final int[] hiveColumnIndexes;
private final boolean[] isPartitionColumn;
private final boolean[] loaded;
private final boolean[] booleans;
private final long[] longs;
private final double[] doubles;
private final byte[][] strings;
private final boolean[] nulls;
private final long totalBytes;
private long completedBytes;
private boolean closed;
private static final Unsafe unsafe;
private static final byte HIVE_EMPTY_STRING_BYTE = (byte) 0xbf;
private static final byte[] EMPTY_STRING = new byte[0];
private static final int SIZE_OF_SHORT = 2;
private static final int SIZE_OF_INT = 4;
private static final int SIZE_OF_LONG = 8;
private static final Set<HiveType> VALID_HIVE_STRING_TYPES = immutableEnumSet(HiveType.BINARY, HiveType.STRING, HiveType.MAP, HiveType.LIST, HiveType.STRUCT);
static {
try {
// fetch theUnsafe object
Field field = Unsafe.class.getDeclaredField("theUnsafe");
field.setAccessible(true);
unsafe = (Unsafe) field.get(null);
if (unsafe == null) {
throw new RuntimeException("Unsafe access not available");
}
// make sure the VM thinks bytes are only one byte wide
if (Unsafe.ARRAY_BYTE_INDEX_SCALE != 1) {
throw new IllegalStateException("Byte array index scale must be 1, but is " + Unsafe.ARRAY_BYTE_INDEX_SCALE);
}
}
catch (Exception e) {
throw new RuntimeException(e);
}
}
public ColumnarBinaryHiveRecordCursor(RecordReader<K, BytesRefArrayWritable> recordReader,
long totalBytes,
Properties splitSchema,
List<HivePartitionKey> partitionKeys,
List<HiveColumnHandle> columns)
{
checkNotNull(recordReader, "recordReader is null");
checkArgument(totalBytes >= 0, "totalBytes is negative");
checkNotNull(splitSchema, "splitSchema is null");
checkNotNull(partitionKeys, "partitionKeys is null");
checkNotNull(columns, "columns is null");
checkArgument(!columns.isEmpty(), "columns is empty");
this.recordReader = recordReader;
this.totalBytes = totalBytes;
this.key = recordReader.createKey();
this.value = recordReader.createValue();
int size = columns.size();
this.names = new String[size];
this.types = new ColumnType[size];
this.hiveTypes = new HiveType[size];
this.fieldInspectors = new ObjectInspector[size];
this.hiveColumnIndexes = new int[size];
this.isPartitionColumn = new boolean[size];
this.loaded = new boolean[size];
this.booleans = new boolean[size];
this.longs = new long[size];
this.doubles = new double[size];
this.strings = new byte[size][];
this.nulls = new boolean[size];
// initialize data columns
try {
StructObjectInspector rowInspector = getTableObjectInspector(splitSchema);
for (int i = 0; i < columns.size(); i++) {
HiveColumnHandle column = columns.get(i);
names[i] = column.getName();
types[i] = column.getType();
hiveTypes[i] = column.getHiveType();
if (!column.isPartitionKey()) {
fieldInspectors[i] = rowInspector.getStructFieldRef(column.getName()).getFieldObjectInspector();
}
hiveColumnIndexes[i] = column.getHiveColumnIndex();
isPartitionColumn[i] = column.isPartitionKey();
}
}
catch (MetaException | SerDeException | RuntimeException e) {
throw Throwables.propagate(e);
}
// parse requested partition columns
Map<String, HivePartitionKey> partitionKeysByName = uniqueIndex(partitionKeys, HivePartitionKey.nameGetter());
for (int columnIndex = 0; columnIndex < columns.size(); columnIndex++) {
HiveColumnHandle column = columns.get(columnIndex);
if (column.isPartitionKey()) {
HivePartitionKey partitionKey = partitionKeysByName.get(column.getName());
checkArgument(partitionKey != null, "Unknown partition key %s", column.getName());
byte[] bytes = partitionKey.getValue().getBytes(Charsets.UTF_8);
switch (types[columnIndex]) {
case BOOLEAN:
if (isTrue(bytes, 0, bytes.length)) {
booleans[columnIndex] = true;
}
else if (isFalse(bytes, 0, bytes.length)) {
booleans[columnIndex] = false;
}
else {
String valueString = new String(bytes, Charsets.UTF_8);
throw new IllegalArgumentException(String.format("Invalid partition value '%s' for BOOLEAN partition key %s", valueString, names[columnIndex]));
}
break;
case LONG:
if (bytes.length == 0) {
throw new IllegalArgumentException(String.format("Invalid partition value '' for BIGINT partition key %s", names[columnIndex]));
}
longs[columnIndex] = parseLong(bytes, 0, bytes.length);
break;
case DOUBLE:
if (bytes.length == 0) {
throw new IllegalArgumentException(String.format("Invalid partition value '' for DOUBLE partition key %s", names[columnIndex]));
}
doubles[columnIndex] = parseDouble(bytes, 0, bytes.length);
break;
case STRING:
strings[columnIndex] = Arrays.copyOf(bytes, bytes.length);
break;
default:
throw new UnsupportedOperationException("Unsupported column type: " + types[columnIndex]);
}
}
}
}
@Override
public long getTotalBytes()
{
return totalBytes;
}
@Override
public long getCompletedBytes()
{
if (!closed) {
updateCompletedBytes();
}
return completedBytes;
}
private void updateCompletedBytes()
{
try {
long newCompletedBytes = (long) (totalBytes * recordReader.getProgress());
completedBytes = min(totalBytes, max(completedBytes, newCompletedBytes));
}
catch (IOException ignored) {
}
}
@Override
public ColumnType getType(int field)
{
return types[field];
}
@Override
public boolean advanceNextPosition()
{
try {
if (closed || !recordReader.next(key, value)) {
close();
return false;
}
// reset loaded flags
// partition keys are already loaded, but everything else is not
System.arraycopy(isPartitionColumn, 0, loaded, 0, isPartitionColumn.length);
// reset null flags
// todo this shouldn't be needed
Arrays.fill(nulls, false);
return true;
}
catch (IOException | RuntimeException e) {
close();
throw Throwables.propagate(e);
}
}
@Override
public boolean getBoolean(int fieldId)
{
checkState(!closed, "Cursor is closed");
validateType(fieldId, ColumnType.BOOLEAN);
if (!loaded[fieldId]) {
parseBooleanColumn(fieldId);
}
return booleans[fieldId];
}
private void parseBooleanColumn(int column)
{
// don't include column number in message because it causes boxing which is expensive here
checkArgument(!isPartitionColumn[column], "Column is a partition key");
loaded[column] = true;
if (hiveColumnIndexes[column] >= value.size()) {
// this partition may contain fewer fields than what's declared in the schema
// this happens when additional columns are added to the hive table after a partition has been created
nulls[column] = true;
}
else {
BytesRefWritable fieldData = value.unCheckedGet(hiveColumnIndexes[column]);
byte[] bytes;
try {
bytes = fieldData.getData();
}
catch (IOException e) {
throw Throwables.propagate(e);
}
int start = fieldData.getStart();
int length = fieldData.getLength();
parseBooleanColumn(column, bytes, start, length);
}
}
private void parseBooleanColumn(int column, byte[] bytes, int start, int length)
{
if (length > 0) {
booleans[column] = bytes[start] != 0;
nulls[column] = false;
}
else {
nulls[column] = true;
}
}
@Override
public long getLong(int fieldId)
{
checkState(!closed, "Cursor is closed");
validateType(fieldId, ColumnType.LONG);
if (!loaded[fieldId]) {
parseLongColumn(fieldId);
}
return longs[fieldId];
}
private void parseLongColumn(int column)
{
// don't include column number in message because it causes boxing which is expensive here
checkArgument(!isPartitionColumn[column], "Column is a partition key");
loaded[column] = true;
if (hiveColumnIndexes[column] >= value.size()) {
// this partition may contain fewer fields than what's declared in the schema
// this happens when additional columns are added to the hive table after a partition has been created
nulls[column] = true;
}
else {
BytesRefWritable fieldData = value.unCheckedGet(hiveColumnIndexes[column]);
byte[] bytes;
try {
bytes = fieldData.getData();
}
catch (IOException e) {
throw Throwables.propagate(e);
}
int start = fieldData.getStart();
int length = fieldData.getLength();
parseLongColumn(column, bytes, start, length);
}
}
private void parseLongColumn(int column, byte[] bytes, int start, int length)
{
if (length == 0) {
nulls[column] = true;
return;
}
nulls[column] = false;
switch (hiveTypes[column]) {
case SHORT:
checkState(length == SIZE_OF_SHORT, "Short should be 2 bytes");
short smallintValue = unsafe.getShort(bytes, (long) Unsafe.ARRAY_BYTE_BASE_OFFSET + start);
longs[column] = Short.reverseBytes(smallintValue);
break;
case TIMESTAMP:
checkState(length >= 1, "Timestamp should be at least 1 byte");
longs[column] = TimestampWritable.getSeconds(bytes, start);
break;
case BYTE:
checkState(length == 1, "Byte should be 1 byte");
longs[column] = bytes[start];
break;
case INT:
checkState(length >= 1, "Int should be at least 1 byte");
if (length == 1) {
longs[column] = bytes[start];
}
else {
int value = 0;
for (int i = 1; i < length; i++) {
value <<= 8;
value |= (bytes[start + i] & 0xFF);
}
longs[column] = WritableUtils.isNegativeVInt(bytes[start]) ? ~value : value;
}
break;
case LONG:
checkState(length >= 1, "Long should be at least 1 byte");
if (length == 1) {
longs[column] = bytes[start];
}
else {
long value = 0;
for (int i = 1; i < length; i++) {
value <<= 8;
value |= (bytes[start + i] & 0xFF);
}
longs[column] = WritableUtils.isNegativeVInt(bytes[start]) ? ~value : value;
}
break;
default:
throw new RuntimeException(String.format("%s is not a valid LONG type", hiveTypes[column]));
}
}
@Override
public double getDouble(int fieldId)
{
checkState(!closed, "Cursor is closed");
validateType(fieldId, ColumnType.DOUBLE);
if (!loaded[fieldId]) {
parseDoubleColumn(fieldId);
}
return doubles[fieldId];
}
private void parseDoubleColumn(int column)
{
// don't include column number in message because it causes boxing which is expensive here
checkArgument(!isPartitionColumn[column], "Column is a partition key");
loaded[column] = true;
if (hiveColumnIndexes[column] >= value.size()) {
// this partition may contain fewer fields than what's declared in the schema
// this happens when additional columns are added to the hive table after a partition has been created
nulls[column] = true;
}
else {
BytesRefWritable fieldData = value.unCheckedGet(hiveColumnIndexes[column]);
byte[] bytes;
try {
bytes = fieldData.getData();
}
catch (IOException e) {
throw Throwables.propagate(e);
}
int start = fieldData.getStart();
int length = fieldData.getLength();
parseDoubleColumn(column, bytes, start, length);
}
}
private void parseDoubleColumn(int column, byte[] bytes, int start, int length)
{
if (length == 0) {
nulls[column] = true;
}
else {
nulls[column] = false;
switch (hiveTypes[column]) {
case FLOAT:
checkState(length == SIZE_OF_INT, "Float should be 4 bytes");
int intBits = unsafe.getInt(bytes, (long) Unsafe.ARRAY_BYTE_BASE_OFFSET + start);
doubles[column] = Float.intBitsToFloat(Integer.reverseBytes(intBits));
break;
case DOUBLE:
checkState(length == SIZE_OF_LONG, "Double should be 8 bytes");
long longBits = unsafe.getLong(bytes, (long) Unsafe.ARRAY_BYTE_BASE_OFFSET + start);
doubles[column] = Double.longBitsToDouble(Long.reverseBytes(longBits));
break;
default:
throw new RuntimeException(String.format("%s is not a valid DOUBLE type", hiveTypes[column]));
}
}
}
@Override
public byte[] getString(int fieldId)
{
checkState(!closed, "Cursor is closed");
validateType(fieldId, ColumnType.STRING);
if (!loaded[fieldId]) {
parseStringColumn(fieldId);
}
return strings[fieldId];
}
private void parseStringColumn(int column)
{
// don't include column number in message because it causes boxing which is expensive here
checkArgument(!isPartitionColumn[column], "Column is a partition key");
loaded[column] = true;
if (hiveColumnIndexes[column] >= value.size()) {
// this partition may contain fewer fields than what's declared in the schema
// this happens when additional columns are added to the hive table after a partition has been created
nulls[column] = true;
}
else {
BytesRefWritable fieldData = value.unCheckedGet(hiveColumnIndexes[column]);
byte[] bytes;
try {
bytes = fieldData.getData();
}
catch (IOException e) {
throw Throwables.propagate(e);
}
int start = fieldData.getStart();
int length = fieldData.getLength();
parseStringColumn(column, bytes, start, length);
}
}
private void parseStringColumn(int column, byte[] bytes, int start, int length)
{
checkState(VALID_HIVE_STRING_TYPES.contains(hiveTypes[column]), "%s is not a valid STRING type", hiveTypes[column]);
if (length == 0) {
nulls[column] = true;
}
else {
nulls[column] = false;
if (hiveTypes[column] == HiveType.MAP || hiveTypes[column] == HiveType.LIST || hiveTypes[column] == HiveType.STRUCT) {
// temporarily special case MAP, LIST, and STRUCT types as strings
// TODO: create a real parser for these complex types when we implement data types
LazyBinaryObject<? extends ObjectInspector> lazyObject = LazyBinaryFactory.createLazyBinaryObject(fieldInspectors[column]);
ByteArrayRef byteArrayRef = new ByteArrayRef();
byteArrayRef.setData(bytes);
lazyObject.init(byteArrayRef, start, length);
strings[column] = SerDeUtils.getJsonBytes(lazyObject.getObject(), fieldInspectors[column]);
}
else {
// TODO: zero length BINARY is not supported. See https://issues.apache.org/jira/browse/HIVE-2483
if (hiveTypes[column] == HiveType.STRING && (length == 1) && bytes[start] == HIVE_EMPTY_STRING_BYTE) {
strings[column] = EMPTY_STRING;
}
else {
strings[column] = Arrays.copyOfRange(bytes, start, start + length);
}
}
}
}
@Override
public boolean isNull(int fieldId)
{
checkState(!closed, "Cursor is closed");
if (!loaded[fieldId]) {
parseColumn(fieldId);
}
return nulls[fieldId];
}
private void parseColumn(int column)
{
switch (types[column]) {
case BOOLEAN:
parseBooleanColumn(column);
break;
case LONG:
parseLongColumn(column);
break;
case DOUBLE:
parseDoubleColumn(column);
break;
case STRING:
parseStringColumn(column);
break;
default:
throw new UnsupportedOperationException("Unsupported column type: " + types[column]);
}
}
private void validateType(int fieldId, ColumnType type)
{
if (types[fieldId] != type) {
// we don't use Preconditions.checkArgument because it requires boxing fieldId, which affects inner loop performance
throw new IllegalArgumentException(String.format("Expected field to be %s, actual %s (field %s)", type, types[fieldId], fieldId));
}
}
@Override
public void close()
{
// some hive input formats are broken and bad things can happen if you close them multiple times
if (closed) {
return;
}
closed = true;
updateCompletedBytes();
try {
recordReader.close();
}
catch (IOException e) {
throw Throwables.propagate(e);
}
}
}