Open GoogleCodeExporter opened 8 years ago
GoogleCodeExporter
commented
8 years ago You say disabling references will cause a stack overflow but from your
stacktrace it looks like you have references enabled.
Does it happen running against trunk?
Can you post an executable code example?Original comment by nathan.s...@gmail.com on 11 Apr 2013 at 4:32
GoogleCodeExporter
commented
8 years ago When references are enabled, as they are in the displayed stack trace, I get
OutOfMemoryError.
Further investigation showed that when disabling references I get an exception
somewhere else and I try to serialize the exception which has a reference to
itself in the cause so that's why I got the stack overflow. So that's a false
flag.
So turning the references off now works (serialization works). But with
references on, it still doesn't work. I'll work at providing a short example.Original comment by rok.lena...@gmail.com on 11 Apr 2013 at 7:53
GoogleCodeExporter
commented
8 years ago I managed to put together an example. Code from this ZIP file produces an
OutOfMemoryError with Kryo 2.21, BUT only on IBM JVMs. It works on Sun JVMs. I
hope some of you have access to IBM JVMs. One big difference between sun and
ibm is that on IBM JVMs "new String(String)" constructor isn't a copy
constructor(doesn't allocate a new char array, but only makes a new String
object (as if calling substring(0)). Thus on IBM JVMs calling "new
String(aHugeString.substring(2,3))" will maintain a reference to the huge char
array of "aHugeString" through the new String (potential memory leak). If that
turns out to be the cause of the problem, use
"".concat(aHugeString.substring(2,3)) to create a string copy with short
internal array.
Hopefully someone here has IBM JVM handy.Original comment by rok.lena...@gmail.com on 22 Apr 2013 at 2:12
Attachments:
GoogleCodeExporter
commented
8 years ago Here's also an image of the local variables and stack trace at the moment of
OutOfMemoryError getting thrown.
Note the difference between size, stash capacity, capacity and keyTable length.Original comment by rok.lena...@gmail.com on 22 Apr 2013 at 2:36
Attachments:
GoogleCodeExporter
commented
8 years ago 
Resolved with a work-around: Made my own ReferenceResolver which uses binary
search over an array (middle road between Map and List resolvers).Original comment by rok.lena...@gmail.com on 21 May 2013 at 1:46
GoogleCodeExporter
commented
8 years ago ObjectMap is used in many projects, I would if there is something wrong there
on an IBM VM or if it is the strings as you mentioned.Original comment by nathan.s...@gmail.com on 11 Jun 2013 at 7:55
GoogleCodeExporter
commented
8 years ago IBM JVM has a number of subtle differences from Sun/Oracle JVM, most notably,
the String(String) constructor is not a copy constructor (though that can't be
the cause in this case). However I do not know what causes ObjectMap to fail. I
am pretty sure my code is correct, as it works as soon as I change JVM from IBM
to Sun. I have the misfortune of having to work with IBM products though so
switching the VM wasn't an option.Original comment by rok.lena...@gmail.com on 12 Jun 2013 at 7:45
GoogleCodeExporter
commented
8 years ago Hi Rok, can you please post your workaround. I'm having same issue as you had.Original comment by mRi...@gmail.com on 4 Oct 2013 at 8:35
GoogleCodeExporter
commented
8 years ago <code>public class BiListReferenceResolver implements
com.esotericsoftware.kryo.ReferenceResolver {
protected Kryo kryo;
protected final ArrayList<Object> readObjects = new ArrayList<Object>();
private int[] writtenObjectsHashes = new int[10];
// objects, then values
private Object[] writtenObjectsAndValues = new Object[20];
private int size = 0;
private int primaryArraySize = 0;
public void setKryo(Kryo kryo) {
this.kryo = kryo;
}
private static int binarySearch(int[] array, int startIndex, int endIndex,
int value) {
int low = startIndex, mid = -1, high = endIndex - 1;
while (low <= high) {
mid = (low + high) >>> 1;
if (value > array[mid]) {
low = mid + 1;
} else if (value == array[mid]) {
return mid;
} else {
high = mid - 1;
}
}
if (mid < 0) {
int insertPoint = endIndex;
for (int index = startIndex; index < endIndex; index++) {
if (value < array[index]) {
insertPoint = index;
}
}
return -insertPoint - 1;
}
return -mid - (value < array[mid] ? 1 : 2);
}
public int addWrittenObject(Object object) {
int id = size;
int hash = System.identityHashCode(object);
int idx = binarySearch(writtenObjectsHashes, 0, primaryArraySize, hash);
if (idx < 0) {
idx = -(idx + 1);
if (primaryArraySize == writtenObjectsHashes.length) {
int[] newHashArray = new int[(writtenObjectsHashes.length * 3) / 2];
System.arraycopy(writtenObjectsHashes, 0, newHashArray, 0, writtenObjectsHashes.length);
writtenObjectsHashes = newHashArray;
Object[] newObjectArray = new Object[newHashArray.length*2];
System.arraycopy(writtenObjectsAndValues, 0, newObjectArray, 0, writtenObjectsAndValues.length);
writtenObjectsAndValues = newObjectArray;
}
for(int i = writtenObjectsHashes.length-1;i > idx ;i--) {
int j = 2*i;
writtenObjectsHashes[i] = writtenObjectsHashes[i-1];
writtenObjectsAndValues[j] = writtenObjectsAndValues[j-2];
writtenObjectsAndValues[j+1] = writtenObjectsAndValues[j-1];
}
writtenObjectsHashes[idx] = hash;
writtenObjectsAndValues[2*idx] = object;
writtenObjectsAndValues[2*idx+1] = id;
primaryArraySize++;
size++;
return id;
} else {
idx = 2 * idx; // objects and values array has bigger indexes
if (writtenObjectsAndValues[idx+1] instanceof Integer) {
// single slot
if (writtenObjectsAndValues[idx] == object) {
return (Integer)writtenObjectsAndValues[idx+1];
} else {
Object[] keys = new Object[4];
int[] values = new int[4];
keys[0] = writtenObjectsAndValues[idx];
values[0] = (Integer)writtenObjectsAndValues[idx+1];
keys[1] = object;
values[1] = id;
writtenObjectsAndValues[idx] = keys;
writtenObjectsAndValues[idx+1] = values;
size++;
return id;
}
} else {
// multiple entry slot
Object[] keys = (Object[])writtenObjectsAndValues[idx];
for(int i = 0;i < keys.length;i++) {
if (keys[i] == object) return ((int[])writtenObjectsAndValues[idx+1])[i];
if (keys[i] == null) {
keys[i] = object;
((int[])writtenObjectsAndValues[idx+1])[i] = id;
size++;
return id;
}
}
// expand
Object[] newKeys = new Object[(keys.length * 3) / 2];
System.arraycopy(keys, 0, newKeys, 0, keys.length);
newKeys[keys.length] = object;
int[] newValues = new int[(keys.length * 3) / 2];
System.arraycopy((int[])writtenObjectsAndValues[idx+1], 0, newValues, 0, keys.length);
writtenObjectsAndValues[idx] = newKeys;
writtenObjectsAndValues[idx+1] = newValues;
size++;
return id;
}
}
}
public int getWrittenId(Object object) {
int hash = System.identityHashCode(object);
int idx = binarySearch(writtenObjectsHashes, 0, primaryArraySize, hash);
if (idx < 0) {
return -1;
} else {
idx = 2 * idx; // objects and values array has bigger indexes
if (writtenObjectsAndValues[idx+1] instanceof Integer) {
// single slot
if (writtenObjectsAndValues[idx] == object) {
return (Integer)writtenObjectsAndValues[idx+1];
} else {
return -1;
}
} else {
// multiple entry slot
Object[] keys = (Object[])writtenObjectsAndValues[idx];
for(int i = 0;i < keys.length;i++) {
if (keys[i] == object) return ((int[])writtenObjectsAndValues[idx+1])[i];
if (keys[i] == null) return -1;
}
return -1;
}
}
}
@SuppressWarnings("rawtypes")
public int nextReadId(Class type) {
int id = readObjects.size();
readObjects.add(null);
return id;
}
public void setReadObject(int id, Object object) {
readObjects.set(id, object);
}
@SuppressWarnings("rawtypes")
public Object getReadObject(Class type, int id) {
return readObjects.get(id);
}
public void reset() {
readObjects.clear();
size = 0;
primaryArraySize = 0;
writtenObjectsAndValues = new Object[20];
writtenObjectsHashes = new int[10];
}
/** Returns false for all primitive wrappers. */
@SuppressWarnings("rawtypes")
public boolean useReferences(Class type) {
return !Util.isWrapperClass(type) && !type.equals(String.class) && !type.equals(Date.class) && !type.equals(BigDecimal.class) && !type.equals(BigInteger.class);
}
public void addReadObject(int id, Object object) {
while (id >= readObjects.size()) readObjects.add(null);
readObjects.set(id, object);
}
}
</code>
Then register it with:
k.setReferenceResolver(new BiListReferenceResolver());Original comment by rok.lena...@gmail.com on 4 Oct 2013 at 10:01
GoogleCodeExporter
commented
8 years ago Scratch that, that resolver is slow as hell. I made a new one just using a
hashmap.
public class ReferenceResolver implements
com.esotericsoftware.kryo.ReferenceResolver {
protected Kryo kryo;
protected Map<Value, Integer> map = new HashMap<Value, Integer>();
protected final ArrayList<Object> readObjects = new ArrayList<Object>();
private static class Value {
private Object val;
private int hash;
public Value(Object val) {
this.val = val;
this.hash = System.identityHashCode(val);
}
@Override
public int hashCode() {
return hash;
}
@Override
public boolean equals(Object obj) {
return val == ((Value) obj).val;
}
}
public void setKryo(Kryo kryo) {
this.kryo = kryo;
}
public int addWrittenObject(Object object) {
Value v = new Value(object);
Integer i = map.get(v);
if (i == null) {
int ret = map.size();
map.put(v, ret);
return ret;
} else {
return i;
}
}
public int getWrittenId(Object object) {
Value v = new Value(object);
Integer i = map.get(v);
if (i == null) {
return -1;
} else {
return i;
}
}
@SuppressWarnings("rawtypes")
public int nextReadId(Class type) {
int id = readObjects.size();
readObjects.add(null);
return id;
}
public void setReadObject(int id, Object object) {
readObjects.set(id, object);
}
@SuppressWarnings("rawtypes")
public Object getReadObject(Class type, int id) {
return readObjects.get(id);
}
public void reset() {
readObjects.clear();
map.clear();
}
/** Returns false for all primitive wrappers. */
@SuppressWarnings("rawtypes")
public boolean useReferences(Class type) {
return !Util.isWrapperClass(type) && !type.equals(String.class) && !type.equals(Date.class) && !type.equals(BigDecimal.class) && !type.equals(BigInteger.class);
}
public void addReadObject(int id, Object object) {
while (id >= readObjects.size()) readObjects.add(null);
readObjects.set(id, object);
}
}
Should be around 8 times faster.Original comment by rok.lena...@gmail.com on 4 Oct 2013 at 11:38
GoogleCodeExporter
commented
8 years ago [deleted comment]Thanks a lot man. Works like a charm!Original comment by mRi...@gmail.com on 4 Oct 2013 at 12:49
Original issue reported on code.google.com by
rok.lena...@gmail.comon 10 Apr 2013 at 1:48Attachments: