Re: Dependencies for partitioned indexes are still a mess

On 2020-Jul-15, Tom Lane wrote:

> Issue #1: "--clean" does not work
> 
> 1. createdb r2
> 2. pg_restore -d r2 regression.dump
> 3. pg_restore --clean -d r2 regression.dump
> 
> pg_restore: while PROCESSING TOC:
> pg_restore: from TOC entry 6606; 1259 35458 INDEX idxpart32_a_idx postgres
> pg_restore: error: could not execute query: ERROR:  cannot drop index public.idxpart32_a_idx because index
public.idxpart3_a_idxrequires it
 
> HINT:  You can drop index public.idxpart3_a_idx instead.
> Command was: DROP INDEX public.idxpart32_a_idx;

I think this problem is just that we're trying to drop a partition index
that's not droppable.  This seems fixed with just leaving the dropStmt
empty, as in the attached.

One issue is that if you previously restored only that particular
partition and its indexes, but not the ATTACH command that would make it
dependent on the parent index, there would not be a DROP command to get
rid of it.  Do we need to be concerned about that case?  I'm inclined to
think not.

> (There seem to be some other problems as well, but most of the 54 complaints
> are related to partitioned indexes/constraints.)

In my run of it there's a good dozen remaining problems, all alike: we
do DROP TYPE widget CASCADE (which works) followed by DROP FUNCTION
public.widget_out(widget), which fails complaining that type widget
doesn't exist.  But in reality the error is innocuous, since that
function was dropped by the DROP TYPE CASCADE anyway.  You could say
that the same thing is happening with these noisy DROP INDEX of index
partitions: the complaints are right in that each partition's DROP INDEX
command doesn't actually work, but the indexes are dropped later anyway,
so the effect is the same.

> Issue #2: parallel restore does not work

Looking.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services