This turned out to be a bit more complex than a simple delete.
QA. This is a memory leak problem. I'm not sure how you generally test these things. The problem appears when using the MODA UDAF, which calculates the average mode of a numeric column with certain tie breakers.

When you can, don't forget to create PRs for the other branches affected by this. 1.4, and possibly 1.2.

Tested in valgrind – no leaks.
Created PR for 1.4

Build verified: 1.5.2-1 (community server b33685)

Run a moda() test on a 1gb dbt3 database:

SELECT MODA(l_extendedprice) FROM lineitem;

Release 1.5.2-1 used less than 1/2 of memory used in 1.4.3-6

1.5.2-1

[centos7:root~]# cat top.log |grep "KiB Mem :" > memStats.txt
[centos7:root~]# head -n 1 memStats.txt
KiB Mem : 41036684 total, 32600892 free, 6431040 used, 2004752 buff/cache
[centos7:root~]# tail -n 1 memStats.txt
KiB Mem : 41036684 total, 32671276 free, 6392944 used, 1972464 buff/cache
[centos7:root~]# sort -k8 -n memStats.txt |tail -n 1
KiB Mem : 41036684 total, 32285488 free, 6776956 used, 1974240 buff/cache

used: 345916

1.4.3-6

[centos7:root~]# head -n 1 memStats.txt
KiB Mem : 5944344 total, 1534740 free, 699560 used, 3710044 buff/cache
[centos7:root~]# tail -n 1 memStats.txt
KiB Mem : 5944344 total, 777252 free, 1487932 used, 3679160 buff/cache
[centos7:root~]# sort -k8 -n memStats.txt |tail -n 1
KiB Mem : 5944344 total, 775656 free, 1489576 used, 3679112 buff/cache

used: 790016