Memory leak like behaviour

Did anyone experience a "memory leak" like behaviour , even in Idle mode.
It seems that we forgot to bound something and ES is just eating all the
memory until it dies.

Here is a memory heap size screenshot from graphite :
is an heap size image from one of the our nodes ( all the other behave the
exact same way ). The cluster is most of the time in idle mode and does
not get any traffic.
big downfall near to 12/05 12PM is where it died , so we had to restart it
again and it started to rise contantly . The somewhere near the end of the
graph we redirected traffic to the nodes so you see the spikes in the
info on the data that is there.
version : 0.19.8
M documents indexed , that take 25GB of data.
shards , 1 replica
4 nodes on 4 x 48GB VM's
the nodes behave the same.
for any help