On Fri, Mar 27, 2020 at 1:29 PM Masahiko Sawada
<masahiko.sawada@2ndquadrant.com> wrote:
>
> On Fri, 27 Mar 2020 at 07:17, Justin Pryzby <pryzby@telsasoft.com> wrote:
> >
> > On Thu, Mar 26, 2020 at 10:04:57AM -0500, Justin Pryzby wrote:
> > > Does that address your comment ?
> >
> > I hope so.
>
> Thank you for updating the patch. I'm concerned a bit about overhead
> of frequently updating and reverting the callback arguments in
> lazy_vacuum_page(). We call that function every time when we vacuum a
> page, but if the table has an index, we actually don't need to update
> the callback arguments in that function. But I hope it's negligible
> since all operation will be performed on memory.
>
Right, it will be a few instructions. I think if there is any
overhead of this, we can easily avoid that by (a) adding a check in
update_vacuum_error_cbarg which tells if the phase is getting changed
or not and if it is not changed, then return, (b) pass additional in
lazy_vacuum_page() to indicate whether we need to change the phase,
(c) just invoke update_vacuum_error_cbarg() in the caller. The
current way appears to be a bit neat than these options, so not sure
if there is an advantage in changing it. Anyway, if we see any
problem with that it is trivial to change it.
--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com