Independent sampling

-- the test is on INNER JOIN to better clarify the impact of independently sample the data

CREATE volatile TABLE t1 (
col  INT
)
PRIMARY INDEX (col)
ON commit preserve ROWS;

CREATE volatile TABLE t2 (
col  INT
)
PRIMARY INDEX (col)
ON commit preserve ROWS;

INSERT INTO t1 (1);
INSERT INTO t1 (2);
INSERT INTO t1 (3);
INSERT INTO t1 (4);
INSERT INTO t1 (5);

INSERT INTO t2 SELECT col FROM t1;

-- it always return one row
SELECT a.col FROM t1  a JOIN t2  b  ON a.col=b.col sample 1;

-- here I first tried to do the testcase with a single table
-- it always return one row (it surprised me, but I'd assume it is an optimization to sample the same set from the same table)
SELECT a.col FROM (SELECT * FROM t1 sample 1) a JOIN (SELECT * FROM t1 sample 1) b  ON a.col=b.col;

-- here we use two different tables, and the sample is working independently
-- it doesn't return always a match
SELECT a.col FROM (SELECT * FROM t1 sample 1) a JOIN (SELECT * FROM t2 sample 1) b  ON a.col=b.col;