Describe the bug
Enabling join dynamic filter pushdown can silently discard a matching row when the filter passes through a ProjectionExec whose output has two columns with the same alias (here a.id and b.id, both named id).
With the same data and SQL, disabling datafusion.optimizer.enable_join_dynamic_filter_pushdown returns one row; enabling it returns zero rows, without an error.
To Reproduce
Run the following SQL in a fresh datafusion-cli session. The two /tmp/ Parquet paths must not already exist.
SET datafusion.optimizer.join_reordering = false;
COPY (SELECT 'a1' AS id, 'x1' AS ty)
TO '/tmp/df_dynamic_filter_projection_a.parquet'
STORED AS PARQUET;
COPY (SELECT 'x1' AS id)
TO '/tmp/df_dynamic_filter_projection_b.parquet'
STORED AS PARQUET;
CREATE EXTERNAL TABLE ta
STORED AS PARQUET
LOCATION '/tmp/df_dynamic_filter_projection_a.parquet';
CREATE EXTERNAL TABLE tb
STORED AS PARQUET
LOCATION '/tmp/df_dynamic_filter_projection_b.parquet';
-- Returns one row: (a1, a1, x1, x1!).
SET datafusion.optimizer.enable_join_dynamic_filter_pushdown = false;
SELECT s.id, a.id, b.id, tag
FROM ta s
JOIN (
SELECT a.id, b.id, a.ty || '!' AS tag
FROM ta a JOIN tb b ON a.ty = b.id
)
ON s.id = a.id;
-- Incorrectly returns zero rows.
SET datafusion.optimizer.enable_join_dynamic_filter_pushdown = true;
SELECT s.id, a.id, b.id, tag
FROM ta s
JOIN (
SELECT a.id, b.id, a.ty || '!' AS tag
FROM ta a JOIN tb b ON a.ty = b.id
)
ON s.id = a.id;
Expected behavior
No response
Additional context
ProjectionExec resolves pushed-down filter columns by output alias instead of by position, so with two outputs named id the filter on b.id is rewritten to a.id. Found while working on #25244; the fix is included in #25259.
Describe the bug
Enabling join dynamic filter pushdown can silently discard a matching row when the filter passes through a
ProjectionExecwhose output has two columns with the same alias (herea.idandb.id, both namedid).With the same data and SQL, disabling
datafusion.optimizer.enable_join_dynamic_filter_pushdownreturns one row; enabling it returns zero rows, without an error.To Reproduce
Run the following SQL in a fresh
datafusion-clisession. The two/tmp/Parquet paths must not already exist.Expected behavior
No response
Additional context
ProjectionExecresolves pushed-down filter columns by output alias instead of by position, so with two outputs namedidthe filter onb.idis rewritten toa.id. Found while working on #25244; the fix is included in #25259.