Re: pg_walinspect - a new extension to get raw WAL data and WAL stats

At Fri, 11 Mar 2022 15:39:13 -0500, Robert Haas <robertmhaas(at)gmail(dot)com> wrote in
> On Thu, Mar 10, 2022 at 9:38 PM Kyotaro Horiguchi
> <horikyota(dot)ntt(at)gmail(dot)com> wrote:
> > It seems to me too rigorous that pg_get_wal_records_info/stats()
> > reject future LSNs as end-LSN and I think WARNING or INFO and stop at
> > the real end-of-WAL is more kind to users. I think the same with the
> > restriction that start and end LSN are required to be different.
>
> In his review just yesterday, Jeff suggested this: "Don't issue
> WARNINGs or other messages for ordinary situations, like when
> pg_get_wal_records_info() hits the end of WAL." I think he's entirely
> right, and I don't think any patch that does otherwise should get

It depends on what we think is the "ordinary" here. If we don't
expect that specified LSN range is not filled-out, the case above is
ordinary and no need for any WARNING nor INFO. I'm fine with that
definition here.

> committed. It is worth remembering that the results of queries are
> often examined by something other than a human being sitting at a psql
> terminal. Any tool that uses this is going to want to understand what
> happened from the result set, not by parsing strings that may show up
> inside warning messages.

Right. I don't think it is right that WARNING is required to evaluate
the result. And I think that the WARNING like 'reached end-of-wal
before end LSN' is a kind that is not required in evaluation of the
result. Since each WAL row contains at least start LSN.

> I think that the right answer here is to have a function that returns
> one row per record parsed, and each row should also include the start
> and end LSN of the record. If for some reason the WAL records return
> start after the specified start LSN (e.g. because we skip over a page
> header) or end before the specified end LSN (e.g. because we reach
> end-of-WAL) the user can figure it out from looking at the LSNs in the
> output rows and comparing them to the LSNs provided as input.

I agree with you here.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center