On Wed, Nov 30, 2022 at 2:44 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Now, we certainly cannot think that these are occurring early in
> postmaster startup. In the wake of 8acd8f869, we should expect
> that there's no further need to call rtld_bind at all in the
> postmaster, but seemingly that's not so. It's very frustrating
> that the backtrace stops where it does :-(. It's also strange
> that we're apparently running with signals enabled whereever
> it is that rtld_bind is getting called from. Could it be that
> sigaction is failing to install the requested signal mask, so
> that one postmaster signal handler is interrupting another?
Add in some code that does sigaction(0, NULL, &mask) to read the
current mask and assert that it's blocked as expected in the handlers?
Start the postmaster in gdb with a break on _rtld_bind to find all the
places that reach it (unexpectedly)?