Re: Force streaming every change in logical decoding

Going with ' logical_decoding_work_mem' seems a reasonable solution, but
since we are mixing
the functionality of developer and production GUC, there is a slight risk
that customer/DBAs may end
up setting it to 0 and forget about it and thus hampering system's
performance.
Have seen many such cases in previous org.

Adding a new developer parameter seems slightly safe, considering we
already have one
such category supported in postgres. It can be on the same line as that of
'force_parallel_mode'.
It will be purely developer GUC, plus if we want to extend something in
future to add/automate
heavier test-cases or any other streaming related dev option, we can extend
the same parameter w/o
disturbing production's one. (see force_parallel_mode=regress for ref).

thanks
Shveta

On Wed, Dec 21, 2022 at 11:25 AM Hayato Kuroda (Fujitsu) <
kuroda(dot)hayato(at)fujitsu(dot)com> wrote:

> Dear Amit,
>
> > The other possibility to achieve what you are saying is that we allow
> > a minimum value of logical_decoding_work_mem as 0 which would mean
> > stream or serialize each change depending on whether the streaming
> > option is enabled.
>
> I understood that logical_decoding_work_mem may double as normal option as
> developer option. I think yours is smarter because we can reduce # of GUCs.
>
> > I think we normally don't allow a minimum value
> > below a certain threshold for other *_work_mem parameters (like
> > maintenance_work_mem, work_mem), so we have followed the same here.
> > And, I think it makes sense from the user's perspective because below
> > a certain threshold it will just add overhead by either writing small
> > changes to the disk or by sending those over the network. However, it
> > can be quite useful for testing/debugging. So, not sure, if we should
> > restrict setting logical_decoding_work_mem below a certain threshold.
> > What do you think?
>
> You mean to say that there is a possibility that users may set a small
> value without deep
> considerations, right? If so, how about using the approach like
> autovacuum_work_mem?
>
> autovacuum_work_mem has a range [-1, MAX_KIROBYTES], and -1 mean that it
> follows
> maintenance_work_mem. If it is set small value like 5KB, its working
> memory is rounded
> up to 1024KB. See check_autovacuum_work_mem().
>
> Based on that, I suggest followings. Can they solve the problem what you
> said?
>
> * If logical_decoding_work_mem is set to 0, all transactions are streamed
> or serialized
> on publisher.
> * If logical_decoding_work_mem is set within [1, 63KB], the value is
> rounded up or ERROR
> is raised.
> * If logical_decoding_work_mem is set greater than or equal to 64KB, the
> set value
> is used.
>
> Best Regards,
> Hayato Kuroda
> FUJITSU LIMITED
> > Amit Kapila.
>