Joe Conway <[EMAIL PROTECTED]> writes: > That's undoubtedly true, and important for the case that isn't memory > constrained (but where I'm already seeing us perform relatively well). > But once we start the machine swapping, runtime goes in the toilet. And > without addressing the memory leak somehow, we will start a machine > swapping significantly earlier than mysql.
I'm not arguing that we don't need to work on the memory usage ... just that I'm not very happy with that particular approach. I wonder whether there is any reasonable way to determine which data structures are responsible for how much space ... in my test I'm seeing MessageContext: 822075440 total in 104 blocks; 4510280 free (1 chunks); 817565160 used ExecutorState: 8024624 total in 3 blocks; 20592 free (12 chunks); 8004032 used so it seems mostly not the executor's fault, but that's not much to go on. regards, tom lane ---------------------------(end of broadcast)--------------------------- TIP 9: In versions below 8.0, the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match