Mayuresh Nirhali <[EMAIL PROTECTED]> writes: > Hi Knut, > > Thanks for the review and trying out my changes. > > Could you tell me with what memory size did you run the repros ?? > I see the OOME for -Xmx16m and do not see it for 64m.
You're right. I used 64 MB, and it did fail eventually. However, I had the repro for DERBY-1142 (with rs.close() removed) running for more than four hours with 16 MB heap, and I didn't get any out of memory error. > My impression is > that there is still some leak happening, though I agree that this is > an edge case. It could be that DERBY-418 has more leaks than DERBY-1142. Since 1142 has a simpler repro, it could be a good idea to focus on that one first. One more comment to your changes: Since it might be perfectly legitimate to have more than 20 open activations, and since there might be a delay between the result set becoming unreachable and it being finalized, the activations vector may be scanned in vain. I think the other solution Dan suggested (setting a flag in GenericLanguageConnectionContext from BaseActivation.markUnused()) is better, since the vector is only scanned when there is an unused activation. Thanks, -- Knut Anders