On Wed, Sep 1, 2021 at 2:39 AM Andres Freund <and...@anarazel.de> wrote: > > Hi, > > On 2021-08-31 18:34:12 +0900, Kyotaro Horiguchi wrote: > > At Tue, 31 Aug 2021 17:14:45 +0900, Masahiko Sawada <sawada.m...@gmail.com> > > wrote in > > > On Tue, Aug 31, 2021 at 2:34 PM Andres Freund <and...@anarazel.de> wrote: > > > > > > > > On 2021-08-31 14:22:39 +0900, Masahiko Sawada wrote: > > > > > You mean to move only the part of sending the message to its own > > > > > before_shmem_exit() callback? or move ReplicationSlotRelease() and > > > > > ReplicationSlotCleanup() from ProcKill() to it? > > > > > > > > The latter. > > > > > > Makes sense. > > > > > > I've attached the patch that moves them to its own > > > before_shmem_exit(). Unless I missed to register the callback it works > > > the same as before except for where to release and clean up the slots. > > > > Is there any reason we need to register the callback dynamically? It > > seems to me what we need to do here is to call the functions at > > before-shmem-exit. > > +1. I'd just add a ReplicationSlotInitialize() to BaseInit().
+1. But BaseInit() is also called by auxiliary processes, which seems not necessary. So isn't it better to add it to InitPostgres() or InitProcess()? Regards, -- Masahiko Sawada EDB: https://www.enterprisedb.com/
