> On Fri, Jan 24, 2020 at 2:13 AM Michael Paquier <michael@paquier.xyz> wrote: > > No, that's not right. I think that it is possible to loop over > > ShmemProtectiveRegion in some cases. And actually, your patch is dead > > wrong because this is some code called by the postmaster and it cannot > > use FATAL. > > Uh, really? I am not aware of such a rule.
I don't think we have ever expressed it as such, but certainly we prefer postmaster to be super robust ... rather live with a some hundred bytes leak rather than have it die and take the whole database service down for what's essentially a fringe bug that has bothered no one in a decade and a half.
Maybe it didn't bother anyone, because the Windows port is much less used.
Anyway, I believe that freeing the memory before returning false, will not bring down the service, changing the patch to LOG, instead of FATAL. The primary error of the patch was to use FATAL.
