group_concat() consumes a great amount of memory

Build tested: 1.0.11-1, 1.0.9-1

Using a VM that's configured with 60 GB memory. I did the following test:

1) create a 1gb dbt3 database
The lineitem table is about 750mb data and 6001215 rows. There are 1500000 unique order keys in the table.

2) execute query
select l_orderkey, group_concat(l_partkey) from lineitem group by l_orderkey;

1500000 rows in set, 1 warning (4 min 13.48 sec)

3) Check cal trace
MariaDB [tpch1]> select calgettrace();
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Desc Mode Table TableOID ReferencedColumns PIO LIO PBE Elapsed Rows
BPS PM orders 3044 (o_custkey,o_orderkey) 1474 1486 0 0.102 1500000
BPS PM lineitem 3092 (l_orderkey,l_partkey) 0 5866 0 215.113 6001215
HJS PM lineitem-orders 3092 - - - - ----- -
TAS UM - - - - - - 244.862 1500000

4) getcalstat

-----------------------------------------------------------+

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
1 row in set (0.01 sec)

ExeMgr's memory utilization was peaked at 40gb. Although group_concat() is known to use more memory, but using 40gb memory to process two columns totaling 44mg of data seems to be excessive.

I also did the same test on 1.0.9-1 and it showed the same behavior.