Thread: FailedAssertion in heap_index_delete_tuples at heapam.c:7220

Hi,

${subject} happened while executing ${attached query} at regresssion
database, using 14dev (commit
d5a83d79c9f9b660a6a5a77afafe146d3c8c6f46) and produced ${attached
stack trace}.

Sadly just loading the regression database and executing this query is
not enough to reproduce. Not sure what else I can do to help with this
one.

-- 
Jaime Casanova
Director de Servicios Profesionales
SystemGuards - Consultores de PostgreSQL

On Tue, Jan 26, 2021 at 10:52 PM Jaime Casanova
<jcasanov@systemguards.com.ec> wrote:
> ${subject} happened while executing ${attached query} at regresssion
> database, using 14dev (commit
> d5a83d79c9f9b660a6a5a77afafe146d3c8c6f46) and produced ${attached
> stack trace}.

I see the bug: gistprunepage() calls
index_compute_xid_horizon_for_tuples() (which ultimately calls the
heapam.c callback for heap_index_delete_tuples()) with an empty array,
which we don't expect. The similar code within _hash_vacuum_one_page()
already only calls index_compute_xid_horizon_for_tuples() when
ndeletable > 0.

The fix is obvious: Bring gistprunepage() in line with
_hash_vacuum_one_page(). I'll go push a fix for that now.

Thanks for the report!
-- 
Peter Geoghegan

On Wed, Jan 27, 2021 at 2:09 AM Peter Geoghegan <pg@bowt.ie> wrote:
>
> On Tue, Jan 26, 2021 at 10:52 PM Jaime Casanova
> <jcasanov@systemguards.com.ec> wrote:
> > ${subject} happened while executing ${attached query} at regresssion
> > database, using 14dev (commit
> > d5a83d79c9f9b660a6a5a77afafe146d3c8c6f46) and produced ${attached
> > stack trace}.
>
> I see the bug: gistprunepage() calls
> index_compute_xid_horizon_for_tuples() (which ultimately calls the
> heapam.c callback for heap_index_delete_tuples()) with an empty array,
> which we don't expect. The similar code within _hash_vacuum_one_page()
> already only calls index_compute_xid_horizon_for_tuples() when
> ndeletable > 0.
>
> The fix is obvious: Bring gistprunepage() in line with
> _hash_vacuum_one_page(). I'll go push a fix for that now.
>

Thanks

-- 
Jaime Casanova
Director de Servicios Profesionales
SystemGuards - Consultores de PostgreSQL