Hello,
I have been playing around with the length less streaming overloads in
JDBC4. In the discussion of DERBY-1471, it was suggested that we forgot
about the layerB streaming in DRDA at the moment and instead implemented
a much simpler approach. When we have what we need, we can improve the
length less overloads.
The planned approach on the client side is to create a LOB and have it
materialize the whole stream in memory to determine the length. The LOB
is then sent to the server by using the existing methods/API, which
require a length argument.
I have seen that the maximum possible size of a byte array is
Integer.MAX_VALUE. However, when I tried this out, I was not able to
create such a big array in all VMs. I got this error message:
java.lang.OutOfMemoryError: Requested array size exceeds VM limit
Does anyone know what's causing these reduced sizes?
Here are the numbers (all Sun VMs):
1.3 2^31 -20
1.4 2^31 -20
1.5, -d32 2^31 -20
1.5, -d64 2^31 -40
1.6, -d32 2^31 -1 <---- as expected
Thus, maximum LOB size in Derby for the new length less overloads is
limited by two factors (seen from the client):
1) Memory on the client (need at least 2xLOB size + overhead)
2) Maximum size of byte array (in range [2^31 -40, 2^31 -1])
Since 1) will be in effect most of the time (still not common with >4G
RAM is it?), 2) will almost never be seen. I don't plan to support LOBs
of 2^31 -1 bytes if the VM don't support that big byte arrays...
Out of curiosity, does anyone know the max limit for byte arrays on
other VMs?
I attached the little program I used, don't forget to set the maximum
heap size high enough.
--
Kristian
public class MaxByteArraySize {
public static void main(String[] args) {
int size = Integer.MAX_VALUE;
byte[] b;
while (true) {
System.out.print(size);
try {
b = new byte[size];
System.out.println(" SUCCESS!");
break;
} catch (Throwable t) {
size--;
System.out.println(" FAILED!");
}
}
}
}
