Matt Allen created AVRO-1881:
--------------------------------
Summary: Avro (Java) Memory Leak when reusing JsonDecoder instance
Key: AVRO-1881
URL: https://issues.apache.org/jira/browse/AVRO-1881
Project: Avro
Issue Type: Bug
Components: java
Affects Versions: 1.8.1
Environment: Ubuntu 15.04
Oracle 1.8.0_91 and OpenJDK 1.8.0_45
Reporter: Matt Allen
{{JsonDecoder}} maintains state for each record decoded, leading to a memory
leak if the same instance is used for multiple inputs. Using
{{JsonDecoder.configure}} to change the input does not correctly clean up the
state stored in {{JsonDecoder.reorderBuffers}}, which leads to an unbounded
number of {{ReorderBuffer}} instances being accumulated. If a new
{{JsonDecoder}} is created for each input there is no memory leak, but it is
significantly more expensive than reusing the same instance.
This problem seems to only occur when the input schema contains a record, which
is consistent with the {{reorderBuffers}} being the source of the leak. My
first look at the {{JsonDecoder}} code leads me to believe that the
{{reorderBuffers}} stack should be empty after a record is fully processed, so
there may be other behavior at play here.
The following is a minimal example which will exhaust a 50MB heap (-Xmx50m)
after about 5.25 million iterations. The first section demonstrates that no
memory leak is encountered when creating a fresh {{JsonDecoder}} instance for
each input.
{code:title=JsonDecoderMemoryLeak.java|borderStyle=solid}
import org.apache.avro.Schema;
import org.apache.avro.io.*;
import org.apache.avro.generic.*;
import java.io.IOException;
public class JsonDecoderMemoryLeak {
public static DecoderFactory decoderFactory = DecoderFactory.get();
public static JsonDecoder createDecoder(String input, Schema schema) throws
IOException {
return decoderFactory.jsonDecoder(schema, input);
}
public static Object decodeAvro(String input, Schema schema, JsonDecoder
decoder) throws IOException {
if (decoder == null) {
decoder = createDecoder(input, schema);
} else {
decoder.configure(input);
}
GenericDatumReader reader = new
GenericDatumReader<GenericRecord>(schema);
return reader.read(null, decoder);
}
public static Schema.Parser parser = new Schema.Parser();
public static Schema schema = parser.parse("{\"name\": \"TestRecord\",
\"type\": \"record\", \"fields\": [{\"name\": \"field1\", \"type\":
\"long\"}]}");
public static String record(long i) {
StringBuilder builder = new StringBuilder("{\"field1\": ");
builder.append(i);
builder.append("}");
return builder.toString();
}
public static void main(String[] args) throws IOException {
// No memory issues when creating a new decoder for each record
System.out.println("Running with fresh JsonDecoder instances for
6000000 iterations");
for(long i = 0; i < 6000000; i++) {
decodeAvro(record(i), schema, null);
}
// Runs out of memory after ~5250000 records
System.out.println("Running with a single reused JsonDecoder instance");
long count = 0;
try {
JsonDecoder decoder = createDecoder(record(0), schema);
while(true) {
decodeAvro(record(count), schema, decoder);
count++;
}
} catch (OutOfMemoryError e) {
System.out.println("Out of memory after " + count + " records");
e.printStackTrace();
}
}
}
{code}
{code:title=Output|borderStyle=solid}
$ java -Xmx50m -jar json-decoder-memory-leak.jar
Running with fresh JsonDecoder instances for 6000000 iterations
Running with a single reused JsonDecoder instance
Out of memory after 5242880 records
java.lang.OutOfMemoryError: Java heap space
at java.util.Arrays.copyOf(Arrays.java:3210)
at java.util.Arrays.copyOf(Arrays.java:3181)
at java.util.Vector.grow(Vector.java:266)
at java.util.Vector.ensureCapacityHelper(Vector.java:246)
at java.util.Vector.addElement(Vector.java:620)
at java.util.Stack.push(Stack.java:67)
at org.apache.avro.io.JsonDecoder.doAction(JsonDecoder.java:487)
at org.apache.avro.io.parsing.Parser.advance(Parser.java:88)
at org.apache.avro.io.JsonDecoder.advance(JsonDecoder.java:139)
at org.apache.avro.io.JsonDecoder.readLong(JsonDecoder.java:178)
at
org.apache.avro.io.ResolvingDecoder.readLong(ResolvingDecoder.java:162)
at
org.apache.avro.generic.GenericDatumReader.readWithoutConversion(GenericDatumReader.java:183)
at
org.apache.avro.generic.GenericDatumReader.read(GenericDatumReader.java:152)
at
org.apache.avro.generic.GenericDatumReader.readField(GenericDatumReader.java:240)
at
org.apache.avro.generic.GenericDatumReader.readRecord(GenericDatumReader.java:230)
at
org.apache.avro.generic.GenericDatumReader.readWithoutConversion(GenericDatumReader.java:174)
at
org.apache.avro.generic.GenericDatumReader.read(GenericDatumReader.java:152)
at
org.apache.avro.generic.GenericDatumReader.read(GenericDatumReader.java:144)
at com.spiceworks.App.decodeAvro(App.java:25)
at com.spiceworks.App.main(App.java:52)
{code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
