Jörg Sonnenberger <[email protected]> writes: > On 7/16/26 12:23 AM, Jason Thorpe wrote: >> >>> On Jul 15, 2026, at 2:58 PM, Mouse <[email protected]> wrote: >>> >>> Except doesn't that open you up to a DoS attack, running the kernel out >>> of memory by creating a zillion semaphores? >> I mean, it’s really the same thing as with file descriptors. > > Yes and no. SYSV semaphores are persistent and independent of both > file descriptors and processes. So even killing processes is not going > to return the memory used by them.
Sure, but if they were fully dynamic, and limit were set to say 2048 semaphors, that would likely not cause real trouble, beyond the kinds of trouble programs can cause now. > PS: the real question would IMO be why PG is still using SYSV > semaphores in first place. That is a good question. My guess is that it's because nobody flipped the define in the pgsql sources to use POSIX semaphores instead on NetBSD and then tested. Do you believe that our POSIX semaphore implementation is 100% solidly correct? (A genuine question - I would hope so, but I have no idea, and I don't hear about them.)
