Re: Re: Add a hint to the "WAL summaries are required" errors

Hi Chao, Manu,

This is registered as
https://www.google.com/url?q=https://commitfest.postgresql.org/patch/7475/&source=gmail&ust=1793900846549000&sa=E
for the current CommitFest. v2 uses Chao's wording and still applies to
master (73a28d2733a). If it looks ready to you, could one of you add
yourself as reviewer there and mark it Ready for Committer?

Regards,
Sai Pranav

On Thu, Oct 08, 2026 07:39 PM, Raja Sai pranav <rajasaipranav0(at)gmail(dot)com>
wrote:

> Hi Chao, Manu,
>
> Thanks both , that's a good correction. I didn't realize the summarizer
> backfills from the oldest WAL still on disk, so "summarization was not
> enabled" isn't really the trigger; it's WAL being recycled before
> summarization catches up. Dropping the conditional causes entirely also
> sidesteps the wal_level=minimal case not fitting them, as Manu pointed out.
>
> v2 attached, using Chao's wording:
>
> HINT: Consider taking a new full backup to use as the basis for future
> incremental backups.
>
> Re-verified the summaries-removed path on master with the new wording.
> Manu, thanks also for independently confirming it on wal_level=minimal and
> that 007 still passes saves me chasing that path myself.
>
> Regards,
> Sai Pranav
>
> On Thu, Oct 8, 2026 at 3:04 PM Manu <manuelreyesbravo(at)gmail(dot)com> wrote:
>
>> Hi Sai,
>>
>> I tried the patch on master (v1 applies cleanly, and 007 still passes
>> 3/3) and reproduced the error on several paths, including the
>> wal_level=minimal case via 007. The hint shows up in all of them.
>>
>> > Would a shorter "Take a new full backup." be preferable?
>>
>> I think there is a case for it. On the minimal path the hint still
>> fires, but its two causes do not really fit: summarize_wal is off there
>> only because minimal requires it -- the actual reason is that the
>> summarizer reaches that WAL and skips it on purpose (fast_forward) --
>> and nothing was removed. And even in the main case the first cause is
>> more a symptom than the condition: with the WAL still on disk, enabling
>> summarize_wal after the full backup does not error, since the summarizer
>> backfills from the oldest WAL present; it only fails once that WAL has
>> been recycled before summarization caught up.
>>
>> So the remedy-only form you suggested reads correctly in all of them.
>> If you would rather keep naming the minimal case, there is existing
>> wording to mirror at xlog.c:6037:
>>
>> Use a backup taken after setting "wal_level" to higher than "minimal".
>>
>> Happy to share the repro script if it helps.
>>
>> Regards,
>> Manu
>>
>