Re: [PATCH] Clarify the behavior of the system when approaching XID wraparound

On Thu, Oct 12, 2023 at 1:10 PM Robert Haas <robertmhaas@gmail.com> wrote:
> On Thu, Oct 12, 2023 at 12:01 PM Peter Geoghegan <pg@bowt.ie> wrote:
> > No objections from me.
>
> Here is a doc-only patch that I think could be back-patched as far as
> emergency mode exists. It combines all of the wording changes to the
> documentation from v1-v3 of the previous version, but without changing
> the message text that is quoted in the documentation, and without
> adding more instances of similar message texts to the documentation,
> and with a bunch of additional hacking by me.

It's a bit weird that we're effectively saying "pay no attention to
that terrible HINT"...but I get it. The simple fact is that the docs
were written in a way that allowed misinformation to catch on -- the
damage that needs to be undone isn't exactly limited to the docs
themselves.

Your choice to not backpatch the changes to the log messages makes a
lot more sense, now that I see that I see the wider context built by
this preparatory patch. Arguably, it would be counterproductive to
pretend that we didn't make this mistake on the backbranches. Better
to own the mistake.

> Some things I changed:
>
> - I made it so that the MXID section refers back to the XID section
> instead of duplicating it, but with a short list of differences.
> - I weakened the existing claim that says you must be a superuser or
> VACUUM definitely won't fix it to say instead that you SHOULD run
> VACUUM as the superuser, because the former is false and the latter is
> true.
> - I made the list of steps for recovering more explicit.
> - I split out the bit about running autovacuum in the affected
> database into a separate step to be performed after VACUUM for
> continued good operation, rather than a necessary ingredient in
> recovery, because it isn't.
> - A bit of other minor rejiggering.

Those all make sense to me.

> I'm not forgetting about the rest of the proposed patch set, or the
> change I proposed earlier. I'm just posting this much now because this
> is how far I got today, and it would be useful to get comments before
> I go further. I think the residual portion of the patch set not
> included in this documentation patch will be quite small, and I think
> that's a good thing, but again, I don't intend to blow that off.

Of course. Your general approach seems wise.

Thanks for working on this. I will be relieved once this is finally
taken care of.

-- 
Peter Geoghegan