On Wed, Oct 16, 2019 at 6:30 PM Alexander Pyhalov <alp@xxxxxxxx> wrote: > I see that at some point several postgresql backends start consuming about 16 GB RAM. If we account for shared_buffers, it meens 4 GB RAM for private backend memory. How can we achieve such numbers? I don't see any long-running (or complex) queries (however, there could be long-running transactions and queries to large partitioned tables). But how could they consume 512* work_mem memory? I'm not sure they ae consuming 512 times the work_memory, I mean there is a whole lot of stuff a process can allocate, and it requires to dig into the process memory map (something I'm not good at!) to understand it. For sure, a single process (backend) can consume one time work_memory per "complex node" in a query plan, that is it can consume multiple times the work_memory value if that is available. Luca
