My replication slave starts allocating all memory once it has reached a certain binlog position. It looks like there is a very large transaction ( affecting >100.000 rows) which ran fine on the master causing this.
I simplified my config already to almost defaults, lowered the pool size and tried playing with max_binlog_cache_size and max_binlog_stmt_cache_size to no avail.

What other info can I supply in order to better diagnose this?

(NB I set MemoryLimit to 70G in limits.conf hence the messages.log showing the OOM at 73400320kB)