Обсуждение: Bug: Missing check_stack_depth() in GRAPH_TABLE rewriter

On Sat, Apr 11, 2026 at 9:28 AM SATYANARAYANA NARLAPURAM
<satyanarlapuram@gmail.com> wrote:
>
> Hi Hackers,
>
> Two recursive functions, generate_setop_from_pathqueries() and
> generate_queries_for_path_pattern_recurse() in GRAPH_TABLE rewriter are missing check_stack_depth() calls. Consider a
propertygraph with  400 edge tables with a 2-hop pattern produce 160,000 path queries and therefore 160,000 recursion
framesin 
> generate_setop_from_pathqueries(). This can easily exceed the OS stack limit for some systems and crashes the
backend.
>
> I am proposing a stop gap fix of checking stack depth by calling check_stack_depth for now but in future we may want
toreduce the recursion depth (for example use a balanced tree to reduce depth from O(N) to O(log N). 
>
> Attached a patch for adding the check_stack_depth() check.
>
> Repro:
>
> CREATE TABLE sv (id int PRIMARY KEY);
>   INSERT INTO sv VALUES (1);
>
>   DO $$
>   BEGIN
>     FOR i IN 1..400 LOOP
>       EXECUTE format(
>         'CREATE TABLE se_%s (id int PRIMARY KEY, src int, dst int)', i);
>     END LOOP;
>   END$$;
>
>   DO $$
>   DECLARE
>     sql text;
>     edges text := '';
>   BEGIN
>     FOR i IN 1..400 LOOP
>       IF i > 1 THEN edges := edges || ', '; END IF;
>       edges := edges || format(
>         'se_%s KEY (id) SOURCE KEY (src) REFERENCES sv (id) '
>         'DESTINATION KEY (dst) REFERENCES sv (id)', i);
>     END LOOP;
>     EXECUTE 'CREATE PROPERTY GRAPH g VERTEX TABLES (sv KEY (id)) '
>          || 'EDGE TABLES (' || edges || ')';
>   END$$;
>
>   SELECT * FROM GRAPH_TABLE(g
>     MATCH (a)-[e1]->(b)-[e2]->(c)
>     COLUMNS(a.id AS a_id))
>   LIMIT 1;
>

Thanks for the report. I could reproduce the segfault on my laptop.
The attached patch fixes it and gives ERROR:  stack depth limit
exceeded.

generate_queries_for_path_pattern_recurse() - has to work in a linear
fashion since the elements need to be processed in an order. Each
permutation of elements produces one query. These queries can be
arranged in a balanced tree as you  suggest OR when constructing the
setop tree we could generate it in divide-and-conquer manner. However,
the tree will be flattened in the planner anyway (See
flatten_simple_union_all() and pull_up_simple_union_all()). Thus the
final planning will require a deeper stack anyway. The code complexity
doesn't seem to be worth it.

I also looked at a few commits that add check_stack_depth() to see if
we add tests for these scenarios. But I didn't find any. So no tests
added with this commit.



--
Best Wishes,
Ashutosh Bapat

On 15.04.26 17:07, Ashutosh Bapat wrote:
> Thanks for the report. I could reproduce the segfault on my laptop.
> The attached patch fixes it and gives ERROR:  stack depth limit
> exceeded.
> 
> generate_queries_for_path_pattern_recurse() - has to work in a linear
> fashion since the elements need to be processed in an order. Each
> permutation of elements produces one query. These queries can be
> arranged in a balanced tree as you  suggest OR when constructing the
> setop tree we could generate it in divide-and-conquer manner. However,
> the tree will be flattened in the planner anyway (See
> flatten_simple_union_all() and pull_up_simple_union_all()). Thus the
> final planning will require a deeper stack anyway. The code complexity
> doesn't seem to be worth it.
> 
> I also looked at a few commits that add check_stack_depth() to see if
> we add tests for these scenarios. But I didn't find any. So no tests
> added with this commit.

committed

(I moved the #include "miscadmin.h" to a more alphabetical position.)

On Fri, Apr 24, 2026 at 11:54 AM Peter Eisentraut <peter@eisentraut.org> wrote:
>
> On 15.04.26 17:07, Ashutosh Bapat wrote:
> > Thanks for the report. I could reproduce the segfault on my laptop.
> > The attached patch fixes it and gives ERROR:  stack depth limit
> > exceeded.
> >
> > generate_queries_for_path_pattern_recurse() - has to work in a linear
> > fashion since the elements need to be processed in an order. Each
> > permutation of elements produces one query. These queries can be
> > arranged in a balanced tree as you  suggest OR when constructing the
> > setop tree we could generate it in divide-and-conquer manner. However,
> > the tree will be flattened in the planner anyway (See
> > flatten_simple_union_all() and pull_up_simple_union_all()). Thus the
> > final planning will require a deeper stack anyway. The code complexity
> > doesn't seem to be worth it.
> >
> > I also looked at a few commits that add check_stack_depth() to see if
> > we add tests for these scenarios. But I didn't find any. So no tests
> > added with this commit.
>
> committed
>

Thanks a lot for committing this and other fixes.

> (I moved the #include "miscadmin.h" to a more alphabetical position.)
>

Didn't notice this. Sorry.

--
Best Wishes,
Ashutosh Bapat