Re: zstd compression for pg_dump

On Fri, Mar 3, 2023 at 10:55 AM Justin Pryzby <pryzby@telsasoft.com> wrote:
> Thanks for looking.  If your zstd library is compiled with thread
> support, could you also try with :workers=N ?  I believe this is working
> correctly, but I'm going to ask for help verifying that...

Unfortunately not (Ubuntu 20.04):

    pg_dump: error: could not set compression parameter: Unsupported parameter

But that lets me review the error! I think these error messages should
say which options caused them.

> It'd be especially useful to test under windows, where pgdump/restore
> use threads instead of forking...  If you have a windows environment but
> not set up for development, I think it's possible to get cirrusci to
> compile a patch for you and then retrieve the binaries provided as an
> "artifact" (credit/blame for this idea should be directed to Thomas
> Munro).

I should be able to do that next week.

> > With this particular dataset, I don't see much improvement with
> > zstd:long.
>
> Yeah.  I this could be because either 1) you already got very good
> comprssion without looking at more data; and/or 2) the neighboring data
> is already very similar, maybe equally or more similar, than the further
> data, from which there's nothing to gain.

What kinds of improvements do you see with your setup? I'm wondering
when we would suggest that people use it.

> I don't want to start exposing lots of fine-granined parameters at this
> point.  In the immediate case, it looks like it may require more than
> just adding another parameter:
>
>               Note: If windowLog is set to larger than 27,
> --long=windowLog or --memory=windowSize needs to be passed to the
> decompressor.

Hm. That would complicate things.

Thanks,
--Jacob