Pig's AvroStorage LOAD removes unicode chars from input - unicode

Question

I am using pig to read avro files and normalize/transform the data before writing back out. The avro files have records of the form:
{
"type" : "record",
"name" : "KeyValuePair",
"namespace" : "org.apache.avro.mapreduce",
"doc" : "A key/value pair",
"fields" : [ {
"name" : "key",
"type" : "string",
"doc" : "The key"
}, {
"name" : "value",
"type" : {
"type" : "map",
"values" : "bytes"
},
"doc" : "The value"
} ]
}
I have used the AvroTools command-line utility in conjunction with jq to dump the first record to JSON:
$ java -jar avro-tools-1.8.1.jar tojson part-m-00000.avro | ./jq --compact-output 'select(.value.pf_v != null)' | head -n 1 | ./jq .
{
"key": "some-record-uuid",
"value": {
"pf_v": "v1\u0003Basic\u0001slcvdr1rw\u001a\u0004v2\u0003DayWatch\u0001slcva2omi\u001a\u0004v3\u0003Performance\u0001slc1vs1v1w1p1g1i\u0004v4\u0003Fundamentals\u0001snlj1erwi\u001a\u0004v5\u0003My Portfolio\u0001svr1dews1b2b3k1k2\u001a\u0004v0\u00035"
}
}
I run the following pig commands:
REGISTER avro-1.8.1.jar
REGISTER json-simple-1.1.1.jar
REGISTER piggybank-0.15.0.jar
REGISTER jackson-core-2.8.6.jar
REGISTER jackson-databind-2.8.6.jar
DEFINE AvroLoader org.apache.pig.piggybank.storage.avro.AvroStorage();
AllRecords = LOAD 'part-m-00000.avro'
USING AvroLoader()
AS (key: chararray, value: map[]);
Records = FILTER AllRecords BY value#'pf_v' is not null;
SmallRecords = LIMIT Records 10;
DUMP SmallRecords;
The corresponding record for the last command above is as follows:
...
(some-record-uuid,[pf_v#v03v1Basicslcviv2DayWatchslcva2omiv3Performanceslc1vs1v1w1p1g1i])
...
As you can see the unicode chars have been removed from the pf_v value. The unicode characters are actually being used as delimiters in these values so I will need them in order to fully parse the records into their desired normalized state. The unicode characters are clearly present in the encoded .avro file (as demonstrated by dumping the file to JSON). Is anybody aware of a way to get AvroStorage to not remove the unicode chars when loading records?
Thank you!
Update:
I have also performed the same operation using Avro's python DataFileReader:
import avro.schema
from avro.datafile import DataFileReader, DataFileWriter
from avro.io import DatumReader, DatumWriter
reader = DataFileReader(open("part-m-00000.avro", "rb"), DatumReader())
for rec in reader:
if 'some-record-uuid' in rec['key']:
print rec
print '--------------------------------------------'
break
reader.close()
This prints a dict with what looks like hex chars substituted for the unicode chars (which is preferable to removing them entirely):
{u'value': {u'pf_v': 'v0\x033\x04v1\x03Basic\x01slcvi\x1a\x04v2\x03DayWatch\x01slcva2omi\x1a\x04v3\x03Performance\x01slc1vs1v1w1p1g1i\x1a'}, u'key': u'some-record-uuid'}