On Mon, Jun 29, 2020 at 10:26 AM Masahiko Sawada
<masahiko.sawada@2ndquadrant.com> wrote:
>
> On Fri, 26 Jun 2020 at 17:53, Amit Kapila <amit.kapila16@gmail.com> wrote:
> >
> > On Fri, Jun 26, 2020 at 11:31 AM Masahiko Sawada
> > <masahiko.sawada@2ndquadrant.com> wrote:
> > >
> > > On Thu, 25 Jun 2020 at 19:35, Amit Kapila <amit.kapila16@gmail.com> wrote:
> > > >
> > > >
> > > > Today, I thought about it again, and if we consider the point that
> > > > logical_decoding_work_mem value doesn’t change much then having the
> > > > stats at slot-level would also allow computing
> > > > logical_decoding_work_mem based on stats. Do you think it is a
> > > > reasonable assumption that users won't change
> > > > logical_decoding_work_mem for different processes (WALSender, etc.)?
> > >
> > > FWIW, if we use logical_decoding_work_mem as a threshold of starting
> > > of sending changes to a subscriber, I think there might be use cases
> > > where the user wants to set different logical_decoding_work_mem values
> > > to different wal senders. For example, setting a lower value to
> > > minimize the latency of synchronous logical replication to a near-site
> > > whereas setting a large value to minimize the amount of data sent to a
> > > far site.
> > >
> >
> > How does setting a large value can minimize the amount of data sent?
> > One possibility is if there are a lot of transaction aborts and
> > transactions are not large enough that they cross
> > logical_decoding_work_mem threshold but such cases shouldn't be many.
>
> Yeah, this is what I meant.
>
> I agree that it would not be a common case that the user sets
> different values for different processes. Based on that assumption, I
> also think having the stats at slot-level is a good idea.
>
Okay.
> But I might
> want to have the reset function.
>
I don't mind but lets fist see how the patch for the basic feature
looks and what is required to implement it? Are you interested in
writing the patch for this work?
--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com