@Override
public ResourceSchema getSchema(String location, Job job) throws IOException {
if (!partitionKeysSet) {
Set<String> keys = getPartitionColumns(location, job);
if (!(keys == null || keys.size() == 0)) {
// re-edit the pigSchema to contain the new partition keys.
ResourceFieldSchema[] fields = pigSchema.getFields();
LOG.debug("Schema: " + Arrays.toString(fields));
ResourceFieldSchema[] newFields = Arrays.copyOf(fields, fields.length + keys.size());
int index = fields.length;
for (String key : keys) {
newFields[index++] = new ResourceFieldSchema(new FieldSchema(key, DataType.CHARARRAY));
}
pigSchema.setFields(newFields);
LOG.debug("Added partition fields: " + keys + " to loader schema");
LOG.debug("Schema is: " + Arrays.toString(newFields));
}
partitionKeysSet = true;
}
return pigSchema;
}
@Test
public void testGetBagSubSchema() throws Exception {
// Define the expected schema.
ResourceFieldSchema[] bagSubFieldSchemas = new ResourceFieldSchema[1];
bagSubFieldSchemas[0] =
new ResourceFieldSchema()
.setName("innertuple")
.setDescription("The tuple in the bag")
.setType(DataType.TUPLE);
ResourceFieldSchema[] innerTupleFieldSchemas = new ResourceFieldSchema[1];
innerTupleFieldSchemas[0] =
new ResourceFieldSchema().setName("innerfield").setType(DataType.CHARARRAY);
bagSubFieldSchemas[0].setSchema(new ResourceSchema().setFields(innerTupleFieldSchemas));
ResourceSchema expected = new ResourceSchema().setFields(bagSubFieldSchemas);
// Get the actual converted schema.
HCatSchema hCatSchema =
new HCatSchema(
Lists.newArrayList(
new HCatFieldSchema("innerLlama", HCatFieldSchema.Type.STRING, null)));
HCatFieldSchema hCatFieldSchema =
new HCatFieldSchema("llama", HCatFieldSchema.Type.ARRAY, hCatSchema, null);
ResourceSchema actual = PigHCatUtil.getBagSubSchema(hCatFieldSchema);
Assert.assertEquals(expected.toString(), actual.toString());
}
/** wrap a pig schema as tuple */
public static ResourceFieldSchema wrapAsTuple(ResourceFieldSchema subFieldSchema)
throws IOException {
ResourceSchema listSchema = new ResourceSchema();
listSchema.setFields(new ResourceFieldSchema[] {subFieldSchema});
ResourceFieldSchema tupleWrapper = new ResourceFieldSchema();
tupleWrapper.setType(DataType.TUPLE);
tupleWrapper.setName(PIG_TUPLE_WRAPPER);
tupleWrapper.setSchema(listSchema);
return tupleWrapper;
}
public void checkSchema(ResourceSchema s) throws IOException {
// Not actually checking schema
// Actually, just storing it to use in the backend
UDFContext udfc = UDFContext.getUDFContext();
Properties p = udfc.getUDFProperties(this.getClass(), new String[] {udfContextSignature});
p.setProperty(SCHEMA_SIGNATURE, s.toString());
}
@Test
public void testGetBagSubSchemaConfigured() throws Exception {
// NOTE: pig-0.8 sets client system properties by actually getting the client
// system properties. Starting in pig-0.9 you must pass the properties in.
// When updating our pig dependency this will need updated.
System.setProperty(HCatConstants.HCAT_PIG_INNER_TUPLE_NAME, "t");
System.setProperty(HCatConstants.HCAT_PIG_INNER_FIELD_NAME, "FIELDNAME_tuple");
UDFContext.getUDFContext().setClientSystemProps(System.getProperties());
// Define the expected schema.
ResourceFieldSchema[] bagSubFieldSchemas = new ResourceFieldSchema[1];
bagSubFieldSchemas[0] =
new ResourceFieldSchema()
.setName("t")
.setDescription("The tuple in the bag")
.setType(DataType.TUPLE);
ResourceFieldSchema[] innerTupleFieldSchemas = new ResourceFieldSchema[1];
innerTupleFieldSchemas[0] =
new ResourceFieldSchema().setName("llama_tuple").setType(DataType.CHARARRAY);
bagSubFieldSchemas[0].setSchema(new ResourceSchema().setFields(innerTupleFieldSchemas));
ResourceSchema expected = new ResourceSchema().setFields(bagSubFieldSchemas);
// Get the actual converted schema.
HCatSchema actualHCatSchema =
new HCatSchema(
Lists.newArrayList(
new HCatFieldSchema("innerLlama", HCatFieldSchema.Type.STRING, null)));
HCatFieldSchema actualHCatFieldSchema =
new HCatFieldSchema("llama", HCatFieldSchema.Type.ARRAY, actualHCatSchema, null);
ResourceSchema actual = PigHCatUtil.getBagSubSchema(actualHCatFieldSchema);
Assert.assertEquals(expected.toString(), actual.toString());
}
/**
* Only read the columns that were requested in the constructor.<br>
*
* @param struct ColumnarStruct
* @param path Path
* @return Tuple
* @throws IOException
*/
private Tuple readColumnarTuple(ColumnarStruct struct, Path path) throws IOException {
int[] columnIndexes = getRequiredColumns();
// the partition keys if any will already be in the UDFContext here.
String[] partitionKeys = getPartitionKeys(null, null);
// only if the path has changed should be run the
if (currentPath == null || !currentPath.equals(path)) {
currentPathPartitionKeyMap =
(partitionKeys == null)
? null
: pathPartitionerHelper.getPathPartitionKeyValues(path.toString());
currentPath = path;
}
// if the partitionColumns is null this value will stop the for loop
// below from trynig to add any partition columns
// that do not exist
int partitionColumnStartIndex = Integer.MAX_VALUE;
if (!(partitionColumns == null || partitionColumns.size() == 0)) {
// partition columns are always appended to the schema fields.
partitionColumnStartIndex = pigSchema.getFields().length;
}
// create tuple with determined previous size
Tuple t = tupleFactory.newTuple(columnIndexes.length);
// read in all columns
for (int i = 0; i < columnIndexes.length; i++) {
int columnIndex = columnIndexes[i];
if (columnIndex < partitionColumnStartIndex) {
Object obj = struct.getField(columnIndex);
Object pigType = HiveRCSchemaUtil.extractPigTypeFromHiveType(obj);
t.set(i, pigType);
} else {
// read the partition columns
// will only be executed if partitionColumns is not null
String key = partitionKeys[columnIndex - partitionColumnStartIndex];
Object value = currentPathPartitionKeyMap.get(key);
t.set(i, value);
}
}
return t;
}
@Override
public ResourceSchema getSchema(String location, Job job) throws IOException {
if (schema != null) {
return schema;
}
final Configuration configuration = job.getConfiguration();
this.initializePhoenixPigConfiguration(location, configuration);
this.schema = PhoenixPigSchemaUtil.getResourceSchema(this.config);
if (LOG.isDebugEnabled()) {
LOG.debug(
String.format(
"Resource Schema generated for location [%s] is [%s]", location, schema.toString()));
}
this.storeInUDFContext(
this.contextSignature, RESOURCE_SCHEMA_SIGNATURE, ObjectSerializer.serialize(schema));
return schema;
}
@Override
public Tuple getNext() throws IOException {
try {
if (!reader.nextKeyValue()) {
return null;
}
final PhoenixRecord phoenixRecord = reader.getCurrentValue();
if (phoenixRecord == null) {
return null;
}
final Tuple tuple = TypeUtil.transformToTuple(phoenixRecord, schema.getFields());
return tuple;
} catch (InterruptedException e) {
int errCode = 6018;
final String errMsg = "Error while reading input";
throw new ExecException(errMsg, errCode, PigException.REMOTE_ENVIRONMENT, e);
}
}
@Override
public void prepareToRead(@SuppressWarnings("rawtypes") RecordReader reader, PigSplit split)
throws IOException {
this.reader = (HiveRCRecordReader) reader;
// check that the required indexes actually exist i.e. the columns that
// should be read.
// assuming this is always defined simplifies the readColumnarTuple
// logic.
int requiredIndexes[] = getRequiredColumns();
if (requiredIndexes == null) {
int fieldLen = pigSchema.getFields().length;
// if any the partition keys should already exist
String[] partitionKeys = getPartitionKeys(null, null);
if (partitionKeys != null) {
fieldLen = partitionKeys.length;
}
requiredIndexes = new int[fieldLen];
for (int i = 0; i < fieldLen; i++) {
requiredIndexes[i] = i;
}
this.requiredColumns = requiredIndexes;
}
try {
serde = new ColumnarSerDe();
serde.initialize(hiveConf, props);
} catch (SerDeException e) {
LOG.error(e.toString(), e);
throw new IOException(e);
}
}
@Override
public void storeSchema(ResourceSchema schema, String location, Job job) throws IOException {
Configuration conf = job.getConfiguration();
DataStorage storage =
new HDataStorage(new Path(location).toUri(), ConfigurationUtil.toProperties(conf));
ElementDescriptor schemaFilePath = storage.asElement(location, schemaFileName);
if (!schemaFilePath.exists() && schema != null) {
try {
new ObjectMapper().writeValue(schemaFilePath.create(), schema);
} catch (JsonGenerationException e) {
log.warn("Unable to write Resource Statistics for " + location);
e.printStackTrace();
} catch (JsonMappingException e) {
log.warn("Unable to write Resource Statistics for " + location);
e.printStackTrace();
}
}
if (printHeaders) {
ElementDescriptor headerFilePath = storage.asElement(location, headerFileName);
if (!headerFilePath.exists()) {
OutputStream os = headerFilePath.create();
try {
String[] names = schema.fieldNames();
String fn;
for (int i = 0; i < names.length; i++) {
fn = ((names[i] == null) ? ("$" + i) : names[i]);
os.write(fn.getBytes("UTF-8"));
if (i < names.length - 1) {
os.write(fieldDel);
} else {
os.write(recordDel);
}
}
} finally {
os.close();
}
}
}
}
/* (non-Javadoc)
* @see org.apache.pig.builtin.PigStorage#putNext(org.apache.pig.data.Tuple)
*
* Given a tuple that corresponds to one record, write
* it out as CSV, converting among Unix/Windows line
* breaks as requested in the instantiation. Also take
* care of escaping field delimiters, double quotes,
* and linebreaks embedded within fields,
*
*/
@Override
public void putNext(Tuple tupleToWrite) throws IOException {
// If WRITE_OUTPUT_HEADER, store a header record with the names of each field
if (storingFirstRecord && headerTreatment == Headers.WRITE_OUTPUT_HEADER && schema != null) {
ArrayList<Object> headerProtoTuple = new ArrayList<Object>();
ResourceFieldSchema[] fields = schema.getFields();
for (ResourceFieldSchema field : fields) {
headerProtoTuple.add(field.getName());
}
super.putNext(tupleMaker.newTuple(headerProtoTuple));
}
storingFirstRecord = false;
ArrayList<Object> mProtoTuple = new ArrayList<Object>();
int embeddedNewlineIndex = -1;
int embeddedCarriageReturn = -1;
String fieldStr = null;
// For good debug messages:
int fieldCounter = -1;
// Do the escaping:
for (Object field : tupleToWrite.getAll()) {
fieldCounter++;
// Substitute a null value with an empty string. See PIG-2470.
if (field == null) {
fieldStr = null;
mProtoTuple.add("");
continue;
}
fieldStr = field.toString();
// Embedded double quotes are replaced by two double quotes:
fieldStr = fieldStr.replaceAll("[\"]", "\"\"");
// If any field delimiters are in the field, or if we did replace
// any double quotes with a pair of double quotes above,
// or if the string includes a newline character (LF:\n:0x0A)
// or includes a carriage return (CR:\r:0x0D)
// and we are to allow newlines in fields,
// then the entire field must be enclosed in double quotes:
embeddedNewlineIndex = fieldStr.indexOf(LINEFEED);
embeddedCarriageReturn = fieldStr.indexOf(CARRIAGE_RETURN);
if ((fieldStr.indexOf(fieldDelimiter) != -1)
|| (fieldStr.indexOf(DOUBLE_QUOTE) != -1)
|| (multilineTreatment == Multiline.YES)
&& (embeddedNewlineIndex != -1 || embeddedCarriageReturn != -1)) {
fieldStr = "\"" + fieldStr + "\"";
}
// If requested: replace any Linefeed-only (^J), with LF-Newline (^M^J),
// This is needed for Excel to recognize a field-internal
// new line:
if ((eolTreatment != Linebreaks.NOCHANGE) && (embeddedNewlineIndex != -1)) {
if (eolTreatment == Linebreaks.WINDOWS) {
loneLFDetector.reset(fieldStr);
loneLFDetector.matches();
fieldStr = loneLFDetector.replaceAll("$1\r\n");
} else if (eolTreatment == Linebreaks.UNIX) {
CRLFDetector.reset(fieldStr);
fieldStr = CRLFDetector.replaceAll("\n");
}
}
mProtoTuple.add(fieldStr);
}
// If Windows line breaks are requested, append
// a newline (0x0D a.k.a. ^M) to the last field
// so that the row termination will end up being
// \r\n, once the superclass' putNext() method
// is done below:
if ((eolTreatment == Linebreaks.WINDOWS) && (fieldStr != null))
mProtoTuple.set(mProtoTuple.size() - 1, fieldStr + "\r");
Tuple resTuple = tupleMaker.newTuple(mProtoTuple);
super.putNext(resTuple);
}
public boolean write(
Object object, ResourceFieldSchema field, Generator generator, boolean writeFieldName) {
byte type = (field != null ? field.getType() : DataType.findType(object));
if (writeFieldName) {
generator.writeFieldName(alias.toES(field.getName()));
}
if (object == null) {
generator.writeNull();
return true;
}
switch (type) {
case DataType.ERROR:
case DataType.UNKNOWN:
return handleUnknown(object, field, generator);
case DataType.NULL:
generator.writeNull();
break;
case DataType.BOOLEAN:
generator.writeBoolean((Boolean) object);
break;
case DataType.INTEGER:
generator.writeNumber(((Number) object).intValue());
break;
case DataType.LONG:
generator.writeNumber(((Number) object).longValue());
break;
case DataType.FLOAT:
generator.writeNumber(((Number) object).floatValue());
break;
case DataType.DOUBLE:
generator.writeNumber(((Number) object).doubleValue());
break;
case DataType.BYTE:
generator.writeNumber((Byte) object);
break;
case DataType.CHARARRAY:
generator.writeString(object.toString());
break;
case DataType.BYTEARRAY:
generator.writeBinary(((DataByteArray) object).get());
break;
// DateTime introduced in Pig 11
case 30: // DataType.DATETIME
generator.writeString(PigUtils.convertDateToES(object));
break;
// DateTime introduced in Pig 12
case 65: // DataType.BIGINTEGER
throw new SerializationException(
"Big integers are not supported by Elasticsearch - consider using a different type (such as string)");
// DateTime introduced in Pig 12
case 70: // DataType.BIGDECIMAL
throw new SerializationException(
"Big decimals are not supported by Elasticsearch - consider using a different type (such as string)");
case DataType.MAP:
ResourceSchema nestedSchema = field.getSchema();
// empty tuple shortcut
if (nestedSchema == null) {
generator.writeBeginObject();
generator.writeEndObject();
break;
}
ResourceFieldSchema[] nestedFields = nestedSchema.getFields();
generator.writeBeginObject();
// Pig maps are actually String -> Object association so we can save the key right away
for (Map.Entry<?, ?> entry : ((Map<?, ?>) object).entrySet()) {
generator.writeFieldName(alias.toES(entry.getKey().toString()));
write(entry.getValue(), nestedFields[0], generator, false);
}
generator.writeEndObject();
break;
case DataType.TUPLE:
nestedSchema = field.getSchema();
// empty tuple shortcut
if (nestedSchema == null) {
generator.writeBeginObject();
generator.writeEndObject();
break;
}
nestedFields = nestedSchema.getFields();
// use getAll instead of get(int) to avoid having to handle Exception...
List<Object> tuples = ((Tuple) object).getAll();
generator.writeBeginObject();
for (int i = 0; i < nestedFields.length; i++) {
String name = nestedFields[i].getName();
// handle schemas without names
name = (StringUtils.hasText(name) ? alias.toES(name) : Integer.toString(i));
generator.writeFieldName(name);
write(tuples.get(i), nestedFields[i], generator, false);
}
generator.writeEndObject();
break;
case DataType.BAG:
nestedSchema = field.getSchema();
// empty tuple shortcut
if (nestedSchema == null) {
generator.writeBeginArray();
generator.writeEndArray();
break;
}
ResourceFieldSchema bagType = nestedSchema.getFields()[0];
generator.writeBeginArray();
for (Tuple tuple : (DataBag) object) {
write(tuple, bagType, generator, false);
}
generator.writeEndArray();
break;
default:
if (writeUnknownTypes) {
return handleUnknown(object, field, generator);
}
return false;
}
return true;
}
