The article ignores the proper modern solution to prevent OOM killing of critical processes - OOM Score Adjust.
Tuning CommitLimit manually is an archaic, imprecise, and error-prone way to handle memory limits, only suitable for single-process workloads that can handle ENOMEM properly. It completely ignores dynamic file page cache memory allocation. You still can get OOM if you get unusually high file activity. On the other hand, under low file activity, it wastes memory on the same page cache, because it can't be reclaimed without memory pressure, and memory pressure can't be created because workload hits ENOMEM earlier. Don't use strict overcommit.