>> It would be great if you could supply a test case which exhibits this
>> behaviour.
>
>I can probably do that relatively easily (though of course subject to the
>behavior of the VM memory manager), though given Oliver Jowett's response
it
>seems pretty obvious what is happening - allocating a 5 MB byte array is
>failing and/or the byte array is constructed in a tight loop allocating
>incrementally larger arrays until the data fits.
It's slightly more "intelligent" - see code...
>Allocating such large arrays is bound to be unreliable in any case, so as
long
>as that is what's happening I fully understand it and there's probably not
>much to be done...
Thats true so long the consumer needs the data fully in RAM, e.g. an
Image-Label or similar. If the consumer is also stream-based it could be
avoided. Unfortunately the backend would still hold the results of the query
fully in memory which is also not always desired...
Guido